Excerpt: GameStop's quarterly filing for the period ended August 1, 2026, submitted to the SEC on September 9, 2026, no longer reports a "Hardware and accessories" line. That category carried the company's console sell-through — the Nintendo Switch 2, Sony PlayStation 5 and Microsoft Xbox Series X named in the annual report filed in March 2026. It is replaced by a new trio: Collectibles, Video Games and Pre-Owned and Refurbished. Six months separate the annual report that still used the old line from the quarter that deleted it, and the prior-period columns reconcile to the cent — hardware-plus-software becomes video-games-plus-pre-owned, exactly. That arithmetic is what makes this a re-carve rather than a relabel, and it removes the last clean public console-hardware sell-through figure available from a US specialty retailer.

A primary-document reading of GameStop's own SEC filings — the 10-K for fiscal 2025, the first- and second-quarter 10-Qs of fiscal 2026, and the earnings release furnished on Form 8-K. Every figure, date and category name below is quoted from those documents.

The short version

There is a single US-listed specialty retailer that, until this quarter, told the public each quarter how much console hardware it had sold. That retailer is GameStop, and on September 9, 2026 it stopped.

The company's 10-Q for the quarter ended August 1, 2026 reports net sales in three categories: Collectibles, Video Games, and Pre-Owned and Refurbished. The three categories it used a quarter earlier — Hardware and accessories, Software, and Collectibles — are gone. The change is disclosed in a single paragraph of Note 3, "Revenue," and prior-period figures are recast to the new presentation.

What matters is not that a line item moved. It is that the new categories are a reallocation of the old ones, and that the reallocation is arithmetically exact against the numbers GameStop itself published in the prior year.

The reconciliation

Here is the comparative column as it was filed in the second-quarter fiscal 2025 10-Q, beside the same comparative column as it appears, recast, in the second-quarter fiscal 2026 10-Q. Both report the three months ended August 2, 2025.

The identity is complete. 592.1 + 152.5 = 744.6, and 494.6 + 250.0 = 744.6. Collectibles carries over unchanged. The six-month column reconciles the same way: 937.4 + 328.1 = 1,265.5 equals 844.8 + 420.7, against a total of 1,704.6 on both sides.

That is the tell. A company that merely renamed a line would leave the old amounts stranded somewhere in the table, or would restate to a different total. GameStop's recast redistributes the exact old balances into the exact new buckets, leaving no remainder and no rounding gap. The console-hardware number, and everything sold alongside it, has been poured into two blended categories and is no longer separable.

What the footnotes say went where

The reallocation is not hidden. Two footnotes define the new Video Games and Pre-Owned and Refurbished lines:

• Video Games "includes new video game hardware, accessories, and software (both physical and digital); digital currency, including platform points and gift cards redeemable for digital games and content; PC gaming products; and extended product warranties sold on new products."
• Pre-Owned and Refurbished "includes pre-owned and refurbished video game hardware, physical video game software, accessories, and consumer electronics, acquired primarily through our customer trade-in programs and extended product warranties sold on pre-owned products."

So new console hardware sits inside Video Games, and trade-in console hardware sits inside Pre-Owned and Refurbished. Neither footnote separates them. A reader can no longer recover the hardware figure the old "Hardware and accessories" line reported on its own.

When exactly it broke

The timing is checkable against consecutive 10-Qs.

• The first-quarter fiscal 2026 10-Q — period ended May 2, 2026, filed June 11, 2026 — still reports exactly the old trio. Its Note 3 table reads: Hardware and accessories $333.7 million (against $345.3 million the year before), Software $152.7 million, Collectibles $348.9 million, total net sales $835.3 million.
• The second-quarter fiscal 2026 10-Q — period ended August 1, 2026, filed September 9, 2026 — has the new trio, with no hardware line.

Three months separate the two filings. The last quarterly document that broke out console hardware was filed in June 2026; the one that stopped was filed in September 2026. The annual report for fiscal 2025, filed March 24, 2026, still described the business in the old terms: "Hardware and accessories. We offer new and pre-owned gaming platforms from the major console manufacturers. The current generation of consoles include the Nintendo Switch 2, Sony PlayStation 5, and Microsoft Xbox Series X."

The company's stated reason

GameStop gives a reason, in Note 3, and it is worth quoting in full because it is more specific than a line-item tidy-up:

"During the second quarter of fiscal 2026, we revised the significant product categories used to disaggregate Net sales to Collectibles, Pre-Owned and Refurbished, and Video Games. The significant categories previously used to disaggregate Net sales were Hardware and accessories, Software, and Collectibles. The revised categories reflect changes in the composition of our business and align the disclosure with how management currently evaluates sales performance. Consumer engagement has shifted toward collectibles and trading cards, which now represent our largest product category. Our pre-owned and refurbished products — sourced primarily through customer trade-ins and including products we repair or refurbish prior to resale — carry economics distinct from sales of new product, and were previously presented within two separate categories. In addition, an increasing proportion of software spending is fulfilled through digital delivery channels, which has narrowed the distinction between the previous Hardware and accessories and Software categories. The change in presentation had no effect on total Net sales, results of operations, financial position, or reportable segments. Prior period amounts have been recast to conform to the current period presentation."

The earnings release furnished on Form 8-K the day before the 10-Q, on September 8, 2026, frames it for investors in one sentence: "Beginning with this quarter, the Company reports net sales in three categories (Collectibles, Video Games, and Pre-Owned and Refurbished), which aligns with how management views and operates the business." Prior-period amounts, it adds, "have been recast to conform to the new presentation."

Two justifications are offered. One is that collectibles and trading cards have become the largest category — a fact stated plainly in the same filing, where Collectibles is 45.1% of net sales against 23.4% a year earlier. The other is that digital delivery has blurred the boundary between hardware and software, so the old split no longer means much. Both are reasons to reorganise a disclosure. Neither is a reason to remove a hardware number that the company had published separately for years.

What the old line looked like, and how long it had existed

GameStop's product-category breakout is not new. The company's annual reports have carried a hardware line in the current form since at least fiscal 2019, and a hardware-and-accessories disclosure in earlier years under different labels.

The fiscal 2019 10-K reports its categories under the heading "Hardware and accessories," and the fiscal 2025 10-K — the last one filed before the change — carries the same heading. Going back further, the fiscal 2013 10-K broke out "New video game hardware" as its own line of the net-sales table, separate from software and accessories. The naming evolved, but the principle did not: for more than a decade, anyone reading GameStop's filings could see console hardware as its own number.

That is what ended in the second quarter of fiscal 2026.

Why a blended category matters to anyone tracking consoles

The console makers themselves have closed their windows one at a time.

Sony stopped publishing quarterly PlayStation hardware unit sales, ending a disclosure it had maintained across the PS4 and PS5 generations. Microsoft stopped reporting Xbox hardware units. Official console sell-through data from the manufacturers is no longer a public series for either platform.

What remained was the retail side, and one name stood out: GameStop sold new console hardware in its stores and told investors, every quarter, what that had rung up. The blended "Video Games" figure that replaced it is larger than hardware alone — it folds in new hardware, all physical and digital software, gift cards and PC products — and the footnotes do not break it back down. The trade-in hardware that used to flow through the same category now sits inside Pre-Owned and Refurbished.

The quarterly read on US console-hardware sell-through at a specialty retailer is therefore no longer publicly available from the company that used to provide it.

The checks against the obvious alternatives

Three explanations were tested and set aside.

It is not an accounting-standard requirement. The segment-disclosure update commonly cited for disclosure changes in this period, ASU 2023-07, is not mentioned in the second-quarter fiscal 2026 10-Q, and the change sits in the revenue-disaggregation note, not in the segment note. This is a presentation choice by the company, made under the same standard it used before.

It is not a standard retail-category reshuffle. Best Buy's most recent annual report contains no video-game-hardware breakout at all; its only relevant revenue line is a blended "Entertainment" category at 7% of revenue. There is no competing US specialty retailer whose filings supply a console-hardware number to replace the one GameStop withdrew.

The recast fact is not itself unreported. The shift toward collectibles — collectibles overtaking video games as the largest category — drew trade coverage when the results landed. What the coverage did not do is reconcile the old and new category tables and establish that the change is an exact redistribution rather than a relabel, which is the arithmetic that shows a hardware disclosure was genuinely removed rather than simply renamed.

What this article is not claiming

It is not a claim that GameStop stopped selling console hardware. The footnotes explicitly place new and pre-owned video game hardware inside the two blended categories. Hardware is sold and is counted; it is simply no longer shown on its own.

It is not a claim that total sales changed. GameStop states the change "had no effect on total Net sales," and the recast comparatives confirm the totals are identical — 972.2 and 1,704.6 on both sides.

It is not an allegation of concealment. A company may reorganise its revenue disclosure for legitimate reasons, and GameStop discloses the change, names the old categories, defines the new ones and recasts prior periods. The point here is what a reader loses, not what the company did wrong.

It is not a claim about why the console makers closed their disclosures. Sony's and Microsoft's reasons for ending hardware-unit reporting are their own; this article only records that their windows are shut and that GameStop's was the last open retail one.

It is not a claim that no other dataset covers console hardware. Circana and equivalent retail trackers publish console sell-through on a commercial basis. The claim is narrower: that a public, filed quarterly hardware line from a US specialty retailer no longer exists.

Sources

• GameStop Corp. — Form 10-Q, quarterly period ended August 1, 2026, filed September 9, 2026 (acc. 0001326380-26-000055), gme-20260801.htm. Note 3, "Revenue": recast net-sales table (Collectibles $356.3M / Video Games $263.2M / Pre-Owned and Refurbished $170.7M for the three months; $705.2M / $549.8M / $370.5M for the six months); the category-definition footnotes (1)–(3); the verbatim rationale for the revision; MD&A gross-margin discussion ("43.7%, from 29.1%"; Video Games "33.3% … compared to 50.9% … which included the launch of the Nintendo Switch 2").
• GameStop Corp. — Form 10-Q, quarterly period ended May 2, 2026, filed June 11, 2026 (acc. 0001326380-26-000025). Note 3 table under the old categories: Hardware and accessories $333.7M, Software $152.7M, Collectibles $348.9M, total net sales $835.3M.
• GameStop Corp. — Form 10-Q, quarterly period ended August 2, 2025, as originally filed. Note 3 table under the old categories: Hardware and accessories $592.1M, Software $152.5M, Collectibles $227.6M, total net sales $972.2M (three months); $937.4M / $328.1M / $439.1M, total $1,704.6M (six months).
• GameStop Corp. — Form 10-K, fiscal year ended January 31, 2026, filed March 24, 2026 (acc. 0001326380-26-000013). "Merchandise" section: "Hardware and accessories. We offer new and pre-owned gaming platforms from the major console manufacturers. The current generation of consoles include the Nintendo Switch 2, Sony PlayStation 5, and Microsoft Xbox Series X."
• GameStop Corp. — Form 8-K, furnished September 8, 2026 (acc. 0001326380-26-000050), Exhibit 99.1 earnings release. "Beginning with this quarter, the Company reports net sales in three categories (Collectibles, Video Games, and Pre-Owned and Refurbished) … Prior-period amounts have been recast to conform to the new presentation. See 'Sales Mix' in Schedule I."
• GameStop Corp. — Form 10-K, fiscal year ended February 2, 2019, for the "Hardware and accessories" category heading; Form 10-K, fiscal year ended February 1, 2014, whose net-sales table breaks out "New video game hardware" as a separate line ($1,730.0M, 19.1% of total).
• Best Buy Co., Inc. — Form 10-K, fiscal year ended January 31, 2026 (CIK 0000764478). Revenue-category table lists Computing and Mobile Phones, Consumer Electronics, Appliances, Entertainment, Services, Other; no video-game-hardware breakout. "Entertainment" is 7% of revenue in both years.
• SEC full-text search (EDGAR FTS). The contiguous phrase "Collectibles, Pre-Owned and Refurbished, and Video Games" returns GameStop's own two September 2026 filings and token-level matches only; no other issuer in EDGAR uses this category trio.

Method note: the filings were downloaded from SEC EDGAR and read as the filed documents; the category tables, the footnote definitions and the rationale paragraph are quoted verbatim, with HTML entities decoded. The reconciliation is arithmetic performed on the figures GameStop itself published in the two comparative columns, and the identity is stated as exactly what those columns show. The claim that the last public hardware line was withdrawn rests on the contrast between the first-quarter 10-Q (old trio present, filed June 11, 2026) and the second-quarter 10-Q (old trio absent, filed September 9, 2026), both quoted above so a reader can check the change directly. No filing read for this article breaks the blended Video Games category down into hardware and software, and this article does not claim one does.

Excerpt: Nintendo's consolidated results for the year ended March 31, 2026 add one new line to the income statement: "Loss on litigation — ¥6,414 million", against zero the year before. But Nintendo's own interim statements carry no such line. The half-year results (November 4, 2025) and the nine-month results (February 3, 2026) list only "loss on disposal of non-current assets" under extraordinary losses. So the whole charge landed in the January–March quarter and surfaced in the full-year release of May 8, 2026 — a month before France's consumer authority went public with a €35 million penalty against Nintendo of Europe over Joy-Con drift. Nintendo's annual report, filed June 25, 2026, values the same line at "USD 40 million." The French penalty was reported as "$40 million." Nintendo does not say which case the ¥6,414 million is.

An original LostInConsoles reading of Nintendo's own financial statements — the half-year, nine-month and full-year kessan tanshin, the annual securities report, and the separate interim releases — set against the published regulatory record of the French consumer penalty. Every figure, date and line-item name below is quoted from those primary documents.

The short version

There is a specific, checkable shape to the only new extraordinary line item Nintendo has booked in years.

Nintendo reports on a fiscal year running from April 1 to March 31. Through FY2026, it published three statements covering that year's trading before the year closed:

The half-year statement's extraordinary-loss block reads, in full: "Loss on disposal of non-current assets 42 … Total extraordinary losses 42." The nine-month statement: "Loss on disposal of non-current assets 54 … Total extraordinary losses 54." Nothing else. The word litigation does not appear as a line item in either.

Then the full-year statement arrives and the block reads: "Loss on litigation — 6,414 / Loss on disposal of non-current assets 268 / Total extraordinary losses 6,683."

The prior-year column is the tell. Next to the ¥6,414 million, the comparative figure for the year ended March 31, 2025 is a bare dash. Zero. Nintendo had never carried this line before.

Because the six-month and nine-month statements both show a clean zero, the entire charge was recognised in the fiscal fourth quarter — the three months from January 1 to March 31, 2026 — and disclosed only when the full-year figures were released.

What Nintendo itself compares it to

Nintendo's annual securities report for the year, published with the annual report dated June 25, 2026, prints the same income statement with a convenience conversion column. Against the litigation line it gives:

Loss on litigation … 6,414 … 40

That column is headed in USD millions, and Nintendo's stated reasonableness rate in the same document is ¥159 to 1 USD. ¥6,414 million at ¥159 per dollar is $40.3 million. So Nintendo's own filing renders the charge as, in effect, USD 40 million.

It is worth being exact about how little that tells a reader. The annual report gives the figure, the USD equivalent, and nothing else. There is no note naming a counterparty, a jurisdiction, a case or a regulator. The only litigation discussion in the document is a generic risk factor — "Nintendo's operations in Japan and outside of Japan may be subject to litigation, disputes and other legal procedures" — with a one-line mitigation, "Nintendo is taking various measures to reduce the risk of litigation." The document that was filed after the public record already contained a named, quantified Nintendo penalty still declines to connect the two.

The other number that is USD 40 million

On June 8, 2026, France's General Directorate for Competition, Consumer Affairs and Fraud Control (DGCCRF) announced it had imposed a €35 million penalty on Nintendo of Europe over the Joy-Con drift defect on the original Switch. The penalty took the form of a transaction pénale — a negotiated criminal settlement — and Nintendo of Europe accepted it.

Every wire report attached the same dollar figure to it. France 24's headline and copy: a fine of "35 million euros ($40 million)." Nintendo Life: "€35 million (just over $40 million)." Le Monde, which broke it in English, named the subsidiary and the amount; the French consumer-association relay reproduces the DGCCRF press release under the title "35 millions d'euros d'amende prononcés à l'issue d'une enquête de la DGCCRF."

So there are two Nintendo loss figures in the public record, a month apart, both sitting at roughly USD 40 million, and the yen figure sitting in Nintendo's extraordinary-losses block.

The arithmetic, done plainly

Take the disclosed charge and divide it by the disclosed French penalty:

¥6,414,000,000 ÷ €35,000,000 = ¥183.26 per euro.

The euro–yen rate through the middle of 2026 sat in that range. The European Central Bank's reference rate for September 18, 2026 was EUR 1 = JPY 180.94. ¥183.26 is not a stretch; it is an ordinary spot rate for the period.

That is not proof, and this article does not present it as proof. Nintendo does not itemise the line, so any attribution is an inference from three independently published figures that happen to agree: a yen amount reported as USD 40 million, a euro amount reported as $40 million, and an implied exchange rate that is the actual exchange rate. What can be said without inference is narrower and still striking — that a company which has never before carried a litigation-loss line booked ¥6,414 million in a single quarter, and that the largest quantified Nintendo loss announced in the same window was the French consumer penalty.

Why the timing matters, and where the accrual logic points

The sequencing is the part that has not been read carefully.

Nintendo's full-year tanshin was published May 8, 2026. The DGCCRF's announcement was June 8, 2026. The annual securities report was submitted June 25, 2026.

That ordering means the ¥6,414 million was already in accounts closed at March 31, 2026 and published on May 8 — a month before the regulator went public. Under accounting rules a loss is provided for when it is probable and reasonably estimable, not when it is publicly announced. A transaction pénale negotiated between a regulator and a listed subsidiary, on an investigation that had been running for years, is exactly the kind of obligation that becomes probable and estimable before the announcement date. The charge appearing in the year-end accounts, and the public confirmation following weeks later, is consistent with a provision recognised in the January–March quarter against a settlement that had not yet been formally announced.

The DGCCRF's own account of the case supports the timeline as a live matter throughout the period. Its investigation followed a complaint filed by the consumer association UFC-Que Choisir in September 2020, and the conduct it penalised spanned 2018 to 2023. This was not a new dispute in early 2026; it was an old one reaching its conclusion.

The explanation the games press settled on, and what it does not fit

The prevailing reading in games coverage is that the ¥6,414 million is the cost of the Palworld patent lawsuit — Nintendo and The Pokémon Company against Pocketpair in the Tokyo District Court.

That reading has a problem of magnitude and a problem of status.

The magnitude problem: the Palworld suit is a damages claim, and the amount claimed is small. Court records, as reported, put the plaintiffs' combined demand at ¥10 million — 5 million yen to each plaintiff — plus late-payment interest. Ten million yen is roughly $66,000. The Palworld case is about an injunction, not money; the money it could produce is around four orders of magnitude below the line Nintendo booked.

The status problem: as of the relevant period the case was ongoing, not resolved. A company does not ordinarily book a litigation loss for a claim it is still prosecuting, and it certainly does not book a loss four hundred times the amount claimed. Reporting that attributed the charge to Palworld did so while noting that the case "drags on," which is precisely the state in which no loss would be recognised.

One outlet went further than the rest. Gizmochina, on June 9, 2026 — the day after the French announcement — wrote that "recent financial filings from the company show a matching ¥6.4 billion loss on litigation, effectively confirming the settlement on the balance sheet." That is the same connection this article makes, stated in a single sentence in one trade post. What that post did not do, and what the rest of the coverage did not do, is decompose the figure: check the half-year and nine-month statements and find the line absent, which fixes the charge to a single quarter, or reconcile the euro and yen amounts against the disclosed dollar figure. The attribution was asserted; the primary-source trail behind it was not walked.

There is a second named candidate in the record worth stating fairly. Malikie Innovations, an Irish licensing firm holding former BlackBerry patents, sued Nintendo in the Western District of Washington and at the Unified Patent Court's Hamburg Local Division over the Switch family, alongside Key Patent Innovations. Reporting on the UPC branch suggests the parties settled, with infringement claims and revocation counterclaims withdrawn. The Malikie/Key Patent thread is the other publicly available explanation for a Nintendo litigation payment, and it is the one a careful reader should weigh against the French fine.

Here the quarter decomposition does real work. If the charge were a Malikie settlement paid in late 2025 — the UPC withdrawal was reported in the closing months of that year — it would appear in the nine-month statement covering April 1 to December 31, 2025. It does not. The nine-month statement's extraordinary-loss block contains one line, and it is non-current assets. Whatever produced the ¥6,414 million was recognised after December 31, 2025.

What this article is not claiming

It is not a claim that Nintendo has confirmed the charge is the French fine. Nintendo does not itemise the line, and nothing in its filings names France, the DGCCRF, Joy-Con, the euro or the transaction. The attribution here is an inference from a disclosed yen figure, a disclosed dollar equivalent, a disclosed euro penalty and a disclosed exchange rate, and it is offered as an inference, flagged as one, not as a disclosure.

It is not a claim that no part of the charge relates to Malikie or anything else. A single ¥6,414 million line could in principle aggregate more than one matter. The fiscal-quarter analysis constrains when the charge was recognised, not what composed it.

It is not a claim that the Palworld case is worth nothing to Nintendo. It is a claim about what the claim is worth — ¥10 million claimed — and about the direction of the money: that case is Nintendo seeking damages, not paying them.

It is not a claim that Nintendo has ever admitted the French findings. Nintendo told Le Monde that the penalty does not constitute an admission of guilt and reflected only the amicable resolution of proceedings — a position it has maintained since the announcement.

It is not a claim derived from a named source inside Nintendo. No filing read for this article breaks the ¥6,414 million down by case, and no Nintendo statement connects the accounting line to the French penalty.

Sources

• Nintendo Co., Ltd. — half-year results (six months ended September 30, 2025): Consolidated Financial Highlights, dated November 4, 2025 — 251104e.pdf. Extraordinary-loss block lists only "Loss on disposal of non-current assets 42 … Total extraordinary losses 42"; no litigation line.
• Nintendo Co., Ltd. — nine-month results (nine months ended December 31, 2025): Consolidated Financial Highlights, dated February 3, 2026 — 260203e.pdf. Extraordinary-loss block lists only "Loss on disposal of non-current assets 54 … Total extraordinary losses 54"; no litigation line.
• Nintendo Co., Ltd. — full-year results (year ended March 31, 2026): Consolidated Financial Highlights, dated May 8, 2026 — 260508e.pdf. Extraordinary-loss block: "Loss on litigation — 6,414" against a prior-year dash; "Loss on disposal of non-current assets 268"; "Total extraordinary losses 6,683."
• Nintendo Co., Ltd. — annual securities report / annual report for the year ended March 31, 2026, submission date June 25, 2026 — annual2603e.pdf. Consolidated income statement with USD convenience column: "Loss on litigation … 6,414 … 40"; stated convenience rate ¥159 to 1 USD; risk factor "Litigation, etc." quoted above; no case named.
• Nintendo Co., Ltd. — first-quarter FY2027 results (three months ended June 30, 2026): Consolidated Financial Highlights, dated August 6, 2026 — 260806e.pdf. Extraordinary-loss block reverts to a single non-current-asset line; no litigation line, and no continuing charge.
• Direction générale de la concurrence, de la consommation et de la répression des fraudes (DGCCRF) — announcement of the €35 million transaction pénale against Nintendo of Europe over Joy-Con drift, June 8, 2026. Primary press release published as 2026-06-08-CP-DGCCRF-Sanction-Nintendo.pdf; text relayed by the consumer association AFOC under the title "Dysfonctionnements des manettes de la console Nintendo Switch 1 : 35 millions d'euros d'amende prononcés à l'issue d'une enquête de la DGCCRF" (afoc.net).
• Reporting on the penalty, naming the subsidiary and the amount:
  • Le Monde, "Nintendo fined €35 million in France over defect in Switch Joy-Con controllers," June 8, 2026 (Nintendo of Europe named; Nintendo's denial of any admission of guilt quoted).
  • France 24, "Nintendo agrees to 35 mn euro French fine over faulty Switch controllers," June 8, 2026 ("35 million euros ($40 million)").
  • Nintendo Life, "Nintendo Fined €35 Million Over Widespread Joy-Con Defects," June 8, 2026 ("€35 million (just over $40 million)").
  • Clubic, "Joy-Con Drift : Nintendo accepte une amende de 35 millions d'euros en France," June 8, 2026 (the transaction pénale; the administrative finding that Nintendo communicated only from 2020).
  • Mediapart, "Nintendo paie 35 millions d'euros et clôt un litige pour pratique commerciale trompeuse," June 8, 2026 (conduct period 2018–2023; UFC-Que Choisir complaint of September 2020).
• Exchange rate: European Central Bank euro reference rate, EUR 1 = JPY 180.94, September 18, 2026 — ECB euro reference rates, JPY graph.
• The competing explanation, stated for balance:
  • ip fray, "Nintendo, Malikie apparently settle UPC patent infringement dispute," December 2, 2025 — Malikie Innovations' September 2024 suit against Nintendo in the Western District of Washington and two UPC Hamburg Local Division complaints; withdrawal of infringement claims and revocation counterclaims.
  • Mondaq, "Malikie, Together With Key Patent Innovations, Sues Nintendo" — the Western District of Washington action over Switch models, dock, charger and parental-control system.
  • Arctic Invent case database, IPR2026-00005 — Nintendo and Malikie Innovations terminated their inter partes review "before trial" following a confidential settlement.
• The attribution this article is testing:
  • GosuGamers, "Nintendo hit with ¥6.4 billion litigation loss following major patent disputes," May 10, 2026 — "For the first time in recent memory, Nintendo has booked a loss on litigation: ¥6,414 million (roughly US$41 million), compared to absolutely nothing in FY2025."
  • Game Rant, "Nintendo Reports Huge Litigation Losses as Palworld Lawsuit Continues," May 8, 2026.
  • Gizmochina, "Nintendo hit with €35 million fine over Switch Joy-Con drift issue," June 9, 2026 — the one trade post that linked the €35 million penalty to the ¥6.4 billion line, in a single sentence.
• The Palworld claim's magnitude: reported court-record summaries putting the plaintiffs' combined damages demand at ¥10 million (5 million yen to each of Nintendo and The Pokémon Company) plus late-payment interest, in the Tokyo District Court action filed September 18, 2024 — e.g. Automaton and Courtdocket, both citing Games Fray's reading of the court records.

Method note: the three interim statements and the annual securities report were downloaded from Nintendo's investor-relations library and read as the original documents; the extraordinary-loss blocks and the USD convenience column are quoted verbatim. The DGCCRF's own press release could not be fetched directly from economie.gouv.fr during this research — the server declined automated requests — so the penalty, the date and the form of the settlement are taken from the published wire and consumer-association relays named above, which reproduce the DGCCRF's release and its wording. The euro/yen reconciliation is arithmetic performed on two disclosed figures and is labelled an inference throughout; Nintendo's filings do not itemise the charge, and this article does not claim they do. The "absent from the half-year and nine-month statements" finding rests on those two documents' extraordinary-loss blocks, which are quoted in full so a reader can check the absence directly.

Excerpt: The U.S. International Trade Commission has run §337 investigations over game hardware before — the Gamevice cases against Nintendo, which it decided on the merits and dismissed. This one is different in kind. Instituted on March 27, 2025 and titled "Certain Video Game Consoles, Routers and Gateways, and Components Thereof," it treated a console as a wireless-networking patent target. The complainant, AX Wireless, LLC of Austin, Texas, is identified by the defensive patent fund Unified Patents as an NPE and "an entity of IdeaHub" — a patent-monetization platform based in Seoul. The named respondents were Sony Interactive Entertainment and the French router and set-top maker Vantiva; the four asserted patents, U.S. 10,917,272 / 11,646,927 / 11,777,776 / 12,063,134, claim header-field configurations in OFDM (802.11) systems. RPX's case note names the accused Sony product outright: the PlayStation 5 Pro. Vantiva settled first, then Sony, and on December 5, 2025 the Commission closed the entire investigation without ever reaching the merits. The settlement terms are not in the notice. No exclusion order issued. No games outlet appears to have reported any of it — and the campaign did not stop: in February 2026 AX Wireless sued four more companies, and in April 2026 the Patent Office reopened one of the four patents, finding substantial new questions of patentability on every challenged claim.

An original LostInConsoles data investigation, read directly from the Commission's own institution and termination notices, the Federal Register record for Investigation No. 337-TA-1445, RPX's case notes on the AX Wireless campaign, and the defensive patent fund Unified Patents' reexamination postings. Every docket number, party, patent number and date below is quoted from those public records.

The short version

There is a whole genre of console-industry news that the trade press covers well and the gaming press ignores: customs rulings, tariff classifications, certification dockets, patent litigation. This story sits at the far end of that genre. A U.S. International Trade Commission case literally titled around "Video Game Consoles" was filed, litigated, and quietly settled — and the thing being accused was not a controller, a disc drive or a chip-design licence. It was the console's Wi-Fi.

The case is Investigation No. 337-TA-1445, "Certain Video Game Consoles, Routers and Gateways, and Components Thereof." The complainant is AX Wireless, LLC, of Austin, Texas. The respondents named in the notice of investigation were Sony Interactive Entertainment Inc. (Tokyo), Sony Interactive Entertainment LLC (San Mateo, California), Vantiva SA (Paris) and Vantiva USA, LLC (Norcross, Georgia). The patents are U.S. Patent Nos. 10,917,272; 11,646,927; 11,777,776; and 12,063,134.

That title is unusual in a specific way. The Commission's own prior gaming cases — the two Gamevice actions against Nintendo over the Switch's detachable controllers — were captioned "Portable Gaming Console Systems With Attachable Handheld Controllers." Those were about mechanical attachment. This case is captioned around consoles, routers and gateways together, and the patents are about radio standards. The defensive patent fund Unified Patents describes the asserted family as "header field configurations (e.g., header repetition) in Orthogonal Frequency Division Multiplexing (OFDM) communication systems." RPX, which tracks the campaign, describes the products as those "that support Wi-Fi 6 (the IEEE 802.11ax standard)."

What the docket actually says

The institution notice is Federal Register document 2025-05172, published March 27, 2025, at 90 FR 13879-80. It states that the complaint was filed by AX Wireless, LLC of Austin, Texas, that it alleges "violations of section 337 ... in the importation into the United States, the sale for importation, or the sale within the United States after importation of certain video game consoles, routers and gateways, and components thereof," and that the Commission's notice of investigation named four respondents — the two Sony entities and the two Vantiva entities — with the Office of Unfair Import Investigations participating.

The respondent addresses come straight off the notice of investigation as served: Sony Interactive Entertainment Inc., 1-7-1 Konan, Minato-ku, Tokyo; Sony Interactive Entertainment LLC, 2207 Bridgepointe Parkway, San Mateo, California; Vantiva SA, 10 Boulevard De Grenelle, Paris; and Vantiva USA, LLC, 4855 Peachtree Industrial Blvd., Suite 200, Norcross, Georgia.

The complainant's own address on the same notice — 2025 Guadalupe Street, Suite 260, Austin, TX 78705 — is the address on the AX Wireless website, which describes the entity as holding a "patent portfolio" of "over 100 assets in more than 10 countries."

What was actually accused

The Commission's termination notice names the four patents, and RPX's case note for the parallel district-court action names the accused Sony product. From RPX's February 2025 write-up of the campaign:

"The Ideahub, Inc. complainant targets the provision of devices (i.e., game consoles, gateways, and routers) that support 'wireless networking technology': for Sony, the PlayStation 5 Pro console and for Vantiva, the ARRIS Surfboard mAX W161 Tri-Band Mesh Ready Wi-Fi 6 router."

So the console named in the ITC campaign is the PlayStation 5 Pro, and the companion accused product is a Wi-Fi 6 mesh router — the ARRIS Surfboard mAX W161. AX Wireless filed the ITC complaint and, in parallel, a district-court case against Sony in the Eastern District of Texas (4:25-cv-00175), asserting "the same four wireless communications patents from the ITC complaint."

Who AX Wireless is

AX Wireless, LLC is a non-practising entity. Unified Patents, which opposes such campaigns, describes it flatly as "an NPE and entity of IdeaHub." IdeaHub, Inc. is a patent-monetization platform; RPX has described a related entity structure in which a Texas subsidiary of a Korean parent is "wholly owned by Ideahub, Inc. (d/b/a IDEAHUB), a patent monetization firm formed in Delaware ... but based in Seoul, Korea." IdeaHub's own LinkedIn describes it as "a patent monetization platform."

This is not a console maker defending a standard. It is a licensing operation, working a wireless-communications patent family, that happened to point it at a game console — and did so alongside a broad campaign against networking and computing hardware.

How it ended: the settlement nobody can read

Vantiva settled first. The termination notice records that the Commission "previously terminated Vantiva SA and Vantiva USA, LLC based on settlement" by Order No. 20 (August 28, 2025), unreviewed by Commission notice on September 18, 2025.

Sony settled next, and the sequence is documented to the day. Per the Commission's termination notice (Federal Register document 2025-22407, published December 10, 2025):

• On September 25, 2025, the complainant filed an unopposed motion to terminate the investigation in its entirety based on a settlement agreement with Sony.
• On September 26, 2025, it filed a corrected version "that attached the settlement, which was inadvertently not attached to the original motion."
• On September 29, 2025, the Office of Unfair Import Investigations filed a response in support; no other responses were filed.
• Also on September 29, 2025, the presiding ALJ issued Order No. 24, granting the unopposed motion, finding it complied with Commission Rule 210.21(b) and that "termination based on settlement would not be contrary to the public interest."
• No party petitioned for review. The Commission "determined not to review" Order No. 24. The Commission vote took place on December 5, 2025, and the investigation was terminated.

Four months. That is the entire active life of the Sony portion of a case the Commission had instituted in March: a motion, a corrected motion, an OUII response, an initial determination, and no review. The notice does not disclose the settlement terms, and no exclusion order ever issued — there was no merits ruling to appeal, because there was no merits ruling at all.

The 2026 sequel: the campaign moved on, and a defensive fund pushed back

AX Wireless did not wind down. RPX's February 2026 write-up records four new district-court complaints — against ASUSTek (2:26-cv-00086, Eastern District of Texas), D-Link (8:26-cv-00252, Central District of California), JH1-Zhao / TP-Link (8:26-cv-00251, Central District of California) and Ubiquiti (1:26-cv-01174, Northern District of Illinois) — plus a parallel ITC complaint against the same defendants (docket 337-TA-3882) "over the same five wireless communications patents." RPX's summary of the target set is broader than consoles: AX Wireless "targets the provision of a wide array of products, ranging from access points to laptops, that support Wi-Fi 6 (the IEEE 802.11ax standard)."

Unified Patents then attacked the patents themselves at the Patent Office. On February 26, 2026, it "filed two ex parte reexamination proceedings against U.S. Patent 10,917,272 and U.S. Patent 11,777,776, owned by AX Wireless, LLC" — two of the same four patents asserted at the ITC against Sony. Two months later, on April 24, 2026, the Central Reexamination Unit granted the request as to the '776 patent, "finding substantial new questions of patentability on all challenged claims."

Unified's own campaign note lists the defendants across the family: "Sony, TP Link, Ubiquiti, D-Link, Vantiva, and ASUSTek." The console name is the first one on the list.

How far the "first" claim goes

The claim in this article is deliberately narrow: that 337-TA-1445 is the first Commission gaming case whose patents and accused products turn on wireless-networking standards rather than controller hardware. Here is the limit of what was searched and what it shows.

A Federal Register full-text search for console-related §337 notices returns the two Gamevice matters — 337-TA-1111 and 337-TA-1197, both captioned "Portable Gaming Console Systems With Attachable Handheld Controllers" — and this one. Both Gamevice cases were decided on the merits and dismissed: the Commission affirmed a finding of no violation in 337-TA-1111 (October 2019) and, on review in 337-TA-1197, again found no violation of section 337 and terminated (September 2021). Neither turned on a wireless standard; both were about detachable controllers.

A separate full-text search for the string "video game console" across Commission notices returns no other §337 investigation titled around a console. That is the basis for calling 337-TA-1445 the first §337 matter captioned around video game consoles as Wi-Fi targets — not "the first console patent case," and not "the first console case ever."

The silence test

The other half of this story is what did not happen. The settlement terminated a case the Commission had itself titled around video game consoles, two weeks before Christmas, with no press. Searches of gaming-outlet terms for the case number, for "PS5 Pro" plus ITC, and for the AX Wireless campaign return trade-press and patent-analytics coverage (RPX, Unified Patents, Chinese trade-remedy bulletins) — and no reporting from the games press.

That is a bounded, honest negative rather than a proof: an absence of indexed coverage is exactly the kind of claim that is easiest to get wrong. What can be said precisely is that the Commission's own primary documents record the whole sequence, the patent-analytics services record the campaign, and no games outlet surfaced in those searches.

What this is not claiming

It is not a claim that Sony shipped an infringing product. The investigation was terminated on settlement before any initial determination on the merits, and an adverse inference cannot be drawn from a settlement in a §337 case — settling is routine and often cheaper than litigating.

It is not a claim about the settlement terms. The Commission's notice says a settlement was attached to the corrected motion and that termination was granted on that basis. It does not disclose the terms, and neither does any public filing read for this article. Amounts, licences, and whether any were paid are unknown.

It is not a claim that the patents are invalid. Two of the four are under reexamination, and the Central Reexamination Unit found "substantial new questions of patentability" on all challenged claims of the '776 patent. That is a threshold finding that a reexamination will proceed — not a final rejection, and not a judgment.

It is not a claim that the accused console is a Wi-Fi 6 device. RPX names the PlayStation 5 Pro as the accused Sony product and describes AX Wireless's target set as Wi-Fi 6 (802.11ax) products. Unified Patents describes the patents themselves as OFDM header-repetition art it characterises as 802.11n-era. Both descriptions are reported here as their sources state them; which standard a console implements is a hardware fact this article does not independently establish.

It is not a claim about motive or coordination. AX Wireless is described by Unified Patents as "an entity of IdeaHub" and by RPX as associated with "the Ideahub, Inc. complainant." What that relationship entails for licensing strategy is not stated in the records read here.

Sources

• U.S. International Trade Commission — institution: Certain Video Game Consoles, Routers and Gateways, and Components Thereof; Notice of Institution of Investigation, Investigation No. 337-TA-1445, 90 FR 13879-80 — Federal Register doc. 2025-05172, published March 27, 2025; Commission press release er0321_66682 (March 21, 2025); notice of investigation as served, 337_1445_notice03212025sgl.pdf (complainant and respondent names and addresses; four asserted patents).
• U.S. International Trade Commission — termination: Notice of a Commission Determination Not To Review an Initial Determination Terminating the Investigation Based on Settlement; Termination of Investigation, Investigation No. 337-TA-1445 — Federal Register doc. 2025-22407, published December 10, 2025 (Order No. 20 and Order No. 24 dates; September 25/26/29, 2025 motion sequence; December 5, 2025 Commission vote). The Commission's electronic docket is EDIS, edis.usitc.gov.
• Prior console cases (for the "first" claim's precision):
  • 337-TA-1111 — Federal Register doc. 2019-22172 (commission determination finding no violation; termination, October 2019); institution via doc. 2018-09464.
  • 337-TA-1197 — Federal Register doc. 2021-19510 (review in part; no violation; termination, September 2021).
• RPX (patent-litigation intelligence) — campaign case notes:
  • "Ideahub's AX Wireless Returns to Litigation, Both in Texas and Before the ITC" (February 22, 2025) — PlayStation 5 Pro and ARRIS Surfboard mAX W161 named as accused products; parallel Eastern District of Texas case 4:25-cv-00175.
  • "AX Wireless Files Another Round of District Court and ITC Complaints" (February 7, 2026) — four new district-court cases (ASUSTek, D-Link, TP-Link, Ubiquiti) and ITC docket 337-TA-3882; Wi-Fi 6 / 802.11ax target description.
• Unified Patents (defensive aggregator) — reexamination postings:
  • "Two AX Wireless Wi-Fi patents challenged" (March 6, 2026) — ex parte reexaminations against U.S. 10,917,272 and 11,777,776, filed February 26, 2026; patents described as OFDM header-repetition art; defendants across the family listed.
  • "AX Wireless Wi-Fi patent challenge granted" (May 21, 2026) — CRU grant on April 24, 2026, finding substantial new questions of patentability on all challenged claims of U.S. 11,777,776.
• AX Wireless, LLC — complainant's own site, axwireless.net (address 2025 Guadalupe Street, Suite 260, Austin, TX 78705; "over 100 assets in more than 10 countries").
• IdeaHub, Inc. — entity description via RPX/Mondaq reporting on the Ideahub campaign and IdeaHub's LinkedIn profile ("a patent monetization platform").

Method note: the docket numbers, captions, respondent names and addresses, patent numbers, order numbers and dates were read from the Commission's own institution and termination notices and the notice of investigation as served. The accused-product identifications and the Wi-Fi 6 characterisation are quoted from RPX's case notes; the patent characterisation and reexamination facts are quoted from Unified Patents' postings. The "first" claim is scoped to a Federal Register full-text search for console-related §337 notices and is stated with its limits in "How far the 'first' claim goes." The absence of games-press coverage is stated as a bounded negative, not as proof. No settlement amount is asserted, and no inference of infringement is drawn from a settlement.

Excerpt: Under Section 1201 of the Copyright Act, the exemption that lets libraries, archives and museums preserve discontinued video games — and lets ordinary owners restore them once a publisher's servers go dark — has to be re-filed every cycle to stay alive. In the 2021 and 2024 cycles that renewal was filed by institutions: the Software Preservation Network and the Library Copyright Alliance, the organizations that built the record behind the exemption. On the 2027 renewal docket the video-game line carries exactly two petitioners, both private individuals, Dan Elkins and Ken Austin. The Library Copyright Alliance did file in the same docket — for computer programs generally and for motion pictures — and left the video-game line unchecked. Nobody has explained the gap, and the comment window closes on September 28, 2026.

An original LostInConsoles data investigation, read directly off the Copyright Office's Section 1201 docket pages for the tenth triennial rulemaking, the filed renewal petitions themselves, the codified exemption text at 37 CFR 201.40, and the Office's own notice of inquiry. Every checkbox, petitioner name and date below is quoted from those public records.

The short version

The right to keep a dead console game playable is not permanent law. It is a regulatory exemption that the Librarian of Congress grants for a fixed term, and it survives only if somebody files to renew it before the term runs out. The Section 1201 rulemaking runs on a fixed cycle; if a class of work goes unrenewed, the exemption covering it lapses, and that activity becomes a circumvention violation again.

For the video-game preservation class, the filing pattern has been institutional. In 2021 the renewal came from the Software Preservation Network (SPN) and the Library Copyright Alliance (LCA). In 2024 it came from the same two organizations. Those dockets are public and the filer names are on the record.

The tenth triennial rulemaking opened this cycle with a notice of inquiry on June 9, 2026, setting renewals for Monday, August 24, 2026 and comments for Monday, September 28, 2026. When the page listing the filed renewal petitions went up, the line reading "Abandoned Video Games" carried two entries. Both are individuals: Dan Elkins, and Ken Austin. Neither is an institution. Neither is the coalition.

This matters because the exemption covers more than hobbyist tinkering. It authorizes libraries, archives and museums to preserve games whose servers have been switched off, and it authorizes owners of lawfully acquired games to restore playability when authentication or other online dependencies stop working. The two 2027 petitioners each describe that as their own reason for filing. The organizations that litigated and documented the class in the previous two cycles are, on this line, absent.

There is a second thing worth stating precisely, because it is easy to get wrong: the Library Copyright Alliance is not missing from the 2027 docket. It filed two renewal petitions this cycle — one for the preservation of computer programs, one for the preservation of motion pictures. What it did not file was a renewal claiming the video-game class. On its own forms, the video-game box is left unchecked.

So the accurate statement is narrow and checkable: the class that keeps discontinued console games alive has, this cycle, only individual renewal petitioners — while the organizations that built and defended the class renew adjacent classes and skip this one.

What the docket actually says

The tenth triennial proceeding lives at copyright.gov/1201/2027/. The Office's own page states the sequence plainly. The notice of inquiry was published June 9, 2026. "Petitions for renewal of current exemptions and petitions proposing new exemptions were due August 24, 2026." And: "Written comments in response to petitions for renewal must be received no later than 11:59 p.m. Eastern Time on September 28, 2026."

That produces a comment window that runs three months from the notice of inquiry to the deadline — a short runway for the only public input the Office takes on whether the record behind an exemption still holds.

The renewal-petition page, at copyright.gov/1201/2027/petitions/renewal/, lists the filings by name. The preservation-relevant lines, quoted exactly:

• "Renewal Pet. – Preservation – Library Copyright Alliance – Motion-Pictures" → Renewal-Pet-Preservation-Library-Copyright-Alliance-Motion-Pictures.pdf
• "Renewal Pet. – Preservation – Library Copyright Alliance – Computer-Programs" → Renewal-Pet-Preservation-Library-Copyright-Alliance-Computer-Programs.pdf
• "Renewal Pet. – Abandoned Video Games – Dan Elkins" → Renewal-Pet-Abandoned-Video-Games-Dan-Elkins.pdf
• "Renewal Pet. – Abandoned Video Games – Ken Austin" → Renewal-Pet-Abandoned-Video-Games-Ken-Austin.pdf

Two organizations' filings, two individual filings. On the page as posted, the string "video game" appears only in the two individual entries. The Library Copyright Alliance's two entries name motion pictures and computer programs, not video games.

The box that was left unchecked

The renewal form asks a petitioner to tick exactly which current exemptions it seeks to renew. The Office's form for the tenth rulemaking lists the classes from the ninth, including this one, verbatim:

"Video games for which outside server support has been discontinued, to allow individual play by gamers and preservation of games by libraries, archives, and museums (as well as necessary jailbreaking of console computer code for preservation uses only), and discontinued video games that never required server support, for preservation by libraries, archives, and museums"

On the 2027 forms, each class line begins with a checkbox glyph. Reading the extracted petition text directly, the convention is unambiguous: a capital "Q" is an empty box, an "@" is a ticked box.

• Dan Elkins' petition — the video-game class line carries @. Ticked.
• Ken Austin's petition — the video-game class line carries @. Ticked.
• Library Copyright Alliance, computer-programs petition — the video-game class line carries Q. Empty. The @ on that form sits on a different line: "Computer programs other than video games, for the preservation of computer programs and computer program-dependent materials by libraries, archives, and museums."
• Library Copyright Alliance, motion-pictures petition — the video-game class line again carries Q. Empty.

That is the whole finding, and it is a matter of glyphs on the face of the documents rather than interpretation. The two individuals each claimed the video-game class. The organization that had claimed it before did not, this time.

What the two individuals actually wrote

Both items are visible on the filings. Elkins, under "Explanation of Need":

"Owners of lawfully acquired video games continue to lose access to gameplay when copyright owners discontinue external server support that has been tightly coupled with authentication or other functions necessary for operation. The exemption protects more than single-player use on one machine. Its definition of 'local gameplay' includes gameplay on a personal computer or video game console, as well as gameplay among locally connected personal computers or consoles, provided that gameplay does not depend on an online service or facility."

Austin, on the same item:

"I know of no changes to the legal landscape surrounding video games that would obviate the need for this exemption to exist, particularly for individual owners of games with [online rights management] that can no longer authenticate due to defunct online servers, for the purpose of restoring playability. I have games in my personal collection that suffer from this problem. Without the exemption, I would effectively be denied lawful access to these games, despite the fact that I lawfully acquired them."

Both filings are, in substance, adequate to the streamlined-renewal standard the Office applies — which is deliberately modest. As the notice of inquiry puts it, the sole purpose of the streamlined proceeding "is to determine whether petitioners have made a minimal showing that the regulatory record that supported a previously issued exemption remains representative of the current marketplace." A petitioner does not have to re-litigate the class. It has to file and say why it still applies.

That is exactly why the composition of the filing list is the story. Under a minimal-showing standard, the identity of the filer is not a formality — it is the entire evidentiary input. When the filers are two individuals rather than a preservation institution, the record the Office weighs is thinner, whatever the merits.

What the organizations said the last time

In 2024, the same class was renewed by SPN and LCA together, and the petition names the two organizations and their submitter of record on its first page: "I, Kendra Albert, submitter of record, represent the Software Preservation Network ('SPN') and work with the Library Copyright Alliance ('LCA'), two national organizations that preserve digital heritage by making out-of-commerce software accessible in support of research, teaching, and learning."

On that 2024 form the video-game class is ticked — the glyph convention on the ninth-rulemaking form is different (0 for empty, ® for ticked), and the marked box sits on the "Video games for which outside server support has been discontinued" line, above the unticked "Computer programs other than video games" line. The 2021 renewal list, for its part, names "Abandoned Video Games – SPN & LCA" as the filer of that class.

So across the two most recent cycles the record is consistent: the class was carried by the coalition. This cycle, on that line, it is carried by two named individuals.

The tests this story failed before it ran

An absence is the easiest kind of claim to get wrong, so it was checked against the ways it could be false.

Is the institutional filing simply not posted yet? The renewal page is live and returns HTTP 200, and it already lists non-video-game renewals from the Library Copyright Alliance. The video-game entries it does list are the two individual ones. A page that has posted the organizations' other classes has not withheld a video-game filing from them.

Is there a third filer, an individual or an organization, buried further down the roster? The full renewal roster was read, not just the video-game header. Searching the page text for other game, console or video renewals returns nothing beyond the Elkins and Austin entries. There is no third video-game renewal petitioner on this docket.

Did the Library Copyright Alliance stop filing altogether? No — it filed twice, for motion pictures and for computer programs. Its absence is specific to the video-game class, not a wholesale withdrawal.

Was the class instead handled in a proposed new exemption rather than a renewal? The page for newly proposed exemptions returns HTTP 404 as of this writing; new-exemption listings for this cycle are still forthcoming. That is a genuine limit on what can be asserted, and it is stated below rather than papered over.

Was 2018 really an institutional filing? It cannot be shown from the archived pages used here: the Copyright Office's 2018 renewal and proposed-petition URLs now serve a 404 shell, not the roster. The claim in this article is therefore built on the two cycles whose filer names could be read directly, 2021 and 2024 — not on the earlier one.

One further check, on the substance: the two individual filings are not trying to smuggle in new language. A renewal petition "may not... request new language or exempted activities for a current exemption," and both filings track the codified class. As the Office warns, "a petitioner may not use a renewal petition to request new language." Nothing in the two individual petitions does.

What this is not claiming

It is not a claim that the exemption will lapse. The Office can renew a class on the strength of the petitions before it, and the streamlined standard is a low bar. Two individual filings that state a continuing need may well satisfy it. The observation is about who is absent, not about the outcome.

It is not a claim about motive. SPN, LCA, the Video Game History Foundation, the Entertainment Software Association and others have argued this class forcefully in prior cycles, and LCA is plainly still active in this one. Why the video-game line carries no organizational filer this time is not stated in any document read for this article.

It is not a count of support. Comments in favour of renewal are due September 28, 2026 and had not yet been filed when the docket was read; the strength of the record may look materially different once they land. This article describes the renewal-petition list as posted, not the final shape of the proceeding.

It is not a legal opinion. Whether an individual filing is materially weaker than an institutional one is a judgment about the rulemaking, and the Office makes it. The checkable facts are the class, the filers, the boxes, and the dates.

Sources

• U.S. Copyright Office — Section 1201 tenth triennial rulemaking (2027 cycle): proceeding page (notice published June 9, 2026; renewals due August 24, 2026; comments due September 28, 2026); renewal-petition index.
  • Renewal Pet. – Abandoned Video Games – Dan Elkins: Renewal-Pet-Abandoned-Video-Games-Dan-Elkins.pdf
  • Renewal Pet. – Abandoned Video Games – Ken Austin: Renewal-Pet-Abandoned-Video-Games-Ken-Austin.pdf
  • Renewal Pet. – Preservation – Library Copyright Alliance – Computer-Programs: Renewal-Pet-Preservation-Library-Copyright-Alliance-Computer-Programs.pdf
  • Renewal Pet. – Preservation – Library Copyright Alliance – Motion-Pictures: Renewal-Pet-Preservation-Library-Copyright-Alliance-Motion-Pictures.pdf
• Comment submission page for the 2027 cycle (deadline 11:59 p.m. Eastern, September 28, 2026). Docket COLC-2026-0100 on regulations.gov.
• Notice of inquiry: Library of Congress, Copyright Office, Exemptions To Permit Circumvention of Access Controls on Copyrighted Works, 37 CFR Part 201, Docket No. 2026-4, published June 9, 2026 — Federal Register doc. 2026-11545. Text of the streamlined-renewal standard and the filing deadlines quoted above.
• Codified exemption: 37 CFR 201.40(b)(19), "Video games in the form of computer programs embodied in physical or downloaded formats," as amended (89 FR 85446, October 28, 2024); the adjacent 37 CFR 201.40(b)(20) covers "Computer programs, except video games," preserved by libraries, archives and museums — the class the Library Copyright Alliance did renew this cycle. Verified against the electronic Code of Federal Regulations.
• Prior-cycle baselines for the same class:
  • 2024 (ninth triennial) renewal list, filer "Software Preservation Network and Library Copyright Alliance"; the SPN/LCA renewal petition for the abandoned-video-games class (submitter of record Kendra Albert).
  • 2021 (eighth triennial) renewal list, filer "Abandoned Video Games – SPN & LCA."
  • 2018 renewal/proposed-petition URL now returns a 404 page and could not be used as a baseline.

Method note: the renewal roster, the class lists and the checkbox glyphs were read from the Copyright Office's own posted pages and from the text of the filed petition PDFs. The glyph convention on the 2027 forms is Q = empty and @ = ticked, and was verified line-by-line against each petition's own class list; the 2024 form uses a different convention (0 / ®). The quoted class definition and the rationales are verbatim. The connecting argument — that the identity of the renewal filer is materially part of the record under a minimal-showing standard — is LostInConsoles' reading of the Office's stated standard, and is labelled as analysis. Statements about the eventual outcome of the proceeding are not predictions; the Office decides the merits.

Excerpt: In December 2024, 94.5% of the video game consoles America imported came from China. By July 2025 that was down to 7.5%, and Vietnam was shipping 63.7% of them. Twelve months later China is back below 10% and Thailand has quietly become the third-largest source. The monthly customs data behind that swing — never assembled in one place for this industry — shows a supply chain that did not relocate over years, but flipped in a single quarter, between February and May 2025. And a parallel paper trail of CBP origin rulings explains the mechanism: with one exception the country that solders the components onto the circuit board is the country the console is "from" — even when the finished machine is still assembled in China.

An original LostInConsoles data investigation, assembled from the UN Comtrade monthly series for HS 950450 (video game consoles and machines), cross-read against the U.S. Customs and Border Protection CROSS country-of-origin rulings issued between April 2025 and June 2026, and the harmonised-tariff rate structure those rulings now feed into. Every figure below is reproducible from the public Comtrade API; the ruling texts are public documents.

The short version

The console industry's move out of China has been reported as a broad strategic shift, usually with a graphic and a year attached. What the underlying monthly customs data shows is narrower and stranger than that: this was not a drift. It was a step change, and it happened across roughly a hundred days.

China's share of U.S. video-game-console imports ran between 69% and 97% every single month from January 2021 through February 2025 — fifty consecutive months, and not once below 69%. The only interruptions to that run in the whole dataset are in early 2020, when Chinese factories were shut for COVID-19 and Vietnam briefly filled the gap; China then climbed back above 90% by October of that year and stayed there. In other words: the last time America's console imports were not overwhelmingly Chinese, the cause was a pandemic, and the disruption lasted months. This time it has lasted seventeen months and counting.

Then:

For the full year: China went from $5.79bn of U.S. console imports in 2024 to $1.66bn in 2025 — a fall of 71.4%. Vietnam went from $341m to $3.75bn, up 1,001%. Over the same period the total value of America's console imports barely moved: $6.71bn in 2024, $6.73bn in 2025.

That last number is the one that matters. The market did not shrink. It was reassigned.

And the reassignment is still running. Through the first seven months of 2026, China's share has averaged 17.5%. Its single largest month this year was February, at $87m — against Vietnam's $170m in the same month. In every month since March, China has been below a third of the market while Vietnam has held between 38% and 68%.

What the number actually is

The series is HS 950450 — "games; video game consoles and machines, other than those of subheading 9504.30." The exclusion matters: 9504.30 is coin- and token-operated games (arcade cabinets, slot machines), so 950450 is the console shelf itself, not the arcade floor. It is not video-game software, not controllers, not accessories, and not general consumer electronics. When this line moves, it is the machines that moved.

The reporting is the U.S. import series, reporter code 842, taken month by month from the UN Comtrade public preview API. That distinction matters too, and it is the first thing a sceptical reader should demand, because "China's share collapsed" is exactly the kind of claim that can be an artefact of how a shipment is labelled rather than what physically happened.

The two ways this number could be lying

It could be a relabelling. If the console is still fully assembled in China but the paperwork declares a Vietnamese origin, the U.S. import column would move to Vietnam without a single factory changing. In trade-data terms this is the single most common way a "supply chain shift" is faked.

It could be a re-routing through a neighbouring tariff line. If shippers reclassified consoles into an adjacent code to dodge a rate, the China number would fall while China's actual output was unchanged.

Both are testable, and both come back negative.

Relabelling: the mirror-statistics test

If China were still producing the consoles and merely declaring them Vietnamese, then China's own export books would keep showing large U.S.-bound shipments. Countries report their own exports; the two sets of books are independent, and their disagreement — "mirror asymmetry" — is the standard tool for detecting exactly this.

They do not disagree. China's own reported exports of HS 950450 to the United States:

• 2023: $7.35bn
• 2024: $3.93bn
• 2025: $1.51bn

China's own statistics record the collapse as clearly as America's do. The 2025 figures are close enough to be the same event seen from two sides: China reports $1.51bn exported to the U.S., the U.S. reports $1.66bn imported from China. A relabelling would have produced a wide and growing gap between those two numbers. There is no gap to find.

Vietnam's own books tell the same story in the other direction: Vietnam reported $963m of HS 950450 exports to the United States in 2023, the year before the flip.

And the country that is genuinely new shows up hardest of all in its own records. Thailand's reported exports of HS 950450 to the United States were $0.1m in 2024. In 2025 they were $374m. That is not a rounding artefact or a reclassification — it is a country going from nothing on this line to a third of a billion dollars in twelve months, and reporting it about itself.

One asymmetry is worth flagging rather than burying, because it is the one place the two sets of books do diverge: on Thailand, the U.S. records $236m of 2025 imports while Thailand reports $374m of exports to the U.S. That 37% gap is the kind of margin that usually indicates goods routed through a third country, a timing difference between when a shipment leaves and when it is entered, or a value declared differently at each end. It is materially smaller than the $4bn swing being explained here, so it does not disturb the finding — but it is the reason the Thai figure is quoted from both ends rather than one, and it is the part of this series a follow-up should chase.

Re-routing: the neighbouring-code test

If the drop were a classification trick, the value should surface in an adjacent line. Checking every neighbouring code in the same heading:

Nothing absorbed it. 950430 and 950440 are small and did not grow. 950490 — the "everything else in heading 9504" bucket — held China at 67.5%, essentially unchanged, and shrank. The only line that moved violently is the one that names the product. A reclassification would have to move $4bn into one of those columns, and none of them has room for it.

So the shift is real, it is physical, and it is specific to consoles.

The mechanism is written down, and it is about a circuit board

The reason a console counts as Vietnamese is the interesting part, and it is not a matter of guesswork. U.S. Customs and Border Protection rules on it, in writing, on request, and has now done so repeatedly for exactly this hardware.

The pattern in the rulings issued from April 2025 onward is that the printed circuit board assembly — the PCBA — is where the product's essential character lives. The country where the components are soldered onto the bare board is the country of origin. The final assembly, which happens in China, does not change that.

Take the Razer controllers, filed by the company's Asia-Pacific arm in Singapore. For the Wolverine V3 Pro and Wolverine V3 Tournament Edition:

"the PCBA contributes to the main functionality of the finished merchandise… The complex SMT manufacturing production of these PCBAs, which includes the placement of numerous individual components onto a bare PCB, creates functional PCBAs that results in a substantial transformation of the components to produce PCBAs of Vietnamese origin."

"Additionally, the assembly process performed in China would not substantially transform the PCBA of Vietnamese origin into a new and different article of commerce with a name, character, and use distinct from that of the exported good."

— CBP ruling N350654, 18 July 2025

And the same formula, with the country swapped, for the Kishi V3 (N356709, December 2025), the Raiju V3 Pro (N362101, June 2026) and an unnamed controller ruled Thai (N355986, December 2025).

For consoles proper, the reasoning is starker. Valve asked CBP to rule on the Steam Deck OLED. The device's main board — carrying the processor — is built in Thailand, then shipped to China, where the supporting boards are made, the whole thing is installed into a housing, and the firmware is loaded. CBP's finding:

"the main PCBA of Thai origin is not substantially transformed in China into a different article of commerce with a new name, character and use."

— CBP ruling N347135, 24 April 2025

And for the Steam Machine, ruled in March 2026:

"the motherboard, which contains the essential functional components of the console, imparts the character of the finished article."

— CBP ruling N359665, 25 March 2026

Read the two halves of that together. The board is made in Thailand. The console is assembled in China. The console is legally Thai.

That is how "$5.79bn of Chinese consoles" becomes "$1.66bn of Chinese consoles" without anybody moving a screwdriver away from Shenzhen. What moved is where the board is soldered — and the customs record follows the board.

Why it moved then

The rulings tell you the how. They do not tell you the when, and the when is the sharpest thing in this dataset.

The flip is not spread across a year. Look at the sequence: February 2025, China at 81%. March: 31.2%. Whatever produced that, it did not have time to be a factory migration. Building a line, qualifying it, and shifting volume through it takes quarters. A 50-point share move inside five weeks does not look like a build-out. It looks like a decision already taken, activated.

Valve's Steam Deck ruling request is dated 26 March 2025. Razer's Wolverine V3 Pro request: 26 June 2025. Turtle Beach's Stealth Ultra request: 9 June 2025. These are companies filing paperwork to lock in an origin position, and they filed it in the same window the import data turned over. The legal position and the physical shift arrive together, which is what you would expect if both are downstream of the same decision.

The tariff environment those rulings now feed into is documented separately: heading 9903.05.77 (Thailand) and 9903.05.84 (Vietnam) carry "the duty provided in the applicable subheading + 12.5%." For HTS 9504.50 the ordinary rate is Free, so the surcharge lands at full value — 12.5% of $6.73bn is roughly $842m of duty on 2025's import volume, or about $478m if this year's seven-month pace were extended across twelve. The origin ruling is the document that decides which column the duty is computed in. That is why these filings happen.

The second front nobody has priced

Vietnam is the story everyone has. Thailand is the one that is still moving, and it is the more interesting of the two because it started from nothing.

In May 2026 Thailand supplied 24.3% of America's console imports — the country's highest month on record. Thailand's 2025 total was $236m against $54m in 2024.

Thailand is not assembling consoles. Thailand is making boards. That is precisely the layer the CBP rulings assign the origin to — which means the component country and the origin country are the same country, and the tariff consequence is priced against the component supplier rather than the brand. Valve's Steam Deck, Steam Machine and Steam Frame rulings all land on Thailand for that reason. Razer's controller rulings land on Thailand from the Kishi V3 onward.

So the map has quietly acquired a third node. Vietnam assembles; Thailand boards; China still assembles, and increasingly does not count.

What this does not say

Three honest limits.

It is a value series, not a unit count. If the mix shifted toward cheaper devices manufactured elsewhere while premium Chinese-made hardware held steady, the dollar collapse would overstate the physical one. The mirror test argues against a pure accounting effect, but value share and unit share are not the same measurement, and this dataset cannot separate them.

Three months of 2026 are still provisional. Comtrade's own preview flags recent periods as subject to revision; the 2026 figures used here run through July and may be restated. The direction and the scale are stable across the whole series, but the 2026 tail is the least settled part of it.

"Console" here means HTS 9504.50. Handheld PCs, gaming laptops and controllers sit in different lines and are absent from every figure above. The controller rulings are cited because they establish the mechanism, not because their dollar value is inside the 950450 series.

A large part of 2025's Chinese number is likely inventory. The December 2024 spike — $1.10bn in a single month, China at 94.5% — reads like an import surge ahead of a change, and the months that follow bleed down rather than stopping. The series shows the label moving; it cannot tell you how much of the residual is goods in transit versus goods still being made.

The bottom line

America's console supply chain did not leave China gradually. Between February and May 2025, China's share of U.S. console imports went from 81% to 22%, Vietnam's went from 6.7% to 59.8%, and the total value of the market did not change. The mechanism is not a mystery: the CROSS rulings spell it out, over and over, in almost identical language — the board is the product, and the board is not made in China any more.

What the data adds is the part the industry commentary has been missing. It gives the date, the magnitude, and the checkable counter-hypotheses. The relabelling theory fails on the mirror statistics. The reclassification theory fails on the neighbouring codes. What is left is a factory layer that relocated, a component layer that relocated to a different country entirely, and a tariff that followed both of them out.

Sources

• United Nations Comtrade — UN Comtrade Database, HS 950450 (Games; video game consoles and machines, other than those of subheading 9504.30), reporter United States (842), monthly and annual import series by partner country, 2020–2026. Public preview API: comtradeapi.un.org. Annual data and code definitions: comtradeplus.un.org
• U.S. Customs and Border Protection — CROSS rulings (country of origin):
  • N347135 (24 Apr 2025) — Valve Steam Deck OLED; main PCBA of Thai origin not substantially transformed by Chinese final assembly → Thailand. rulings.cbp.gov/ruling/N347135
  • N349996 (1 Jul 2025) — Turtle Beach Stealth Ultra Wireless XB/PC controller; eight PCBAs assembled and tested in Vietnam → Vietnam. rulings.cbp.gov/ruling/N349996
  • N350654 (18 Jul 2025) — Razer Wolverine V3 Pro / V3 Tournament Edition; Vietnam SMT → Vietnam. rulings.cbp.gov/ruling/N350654
  • N355069 (6 Nov 2025) — Razer Wolverine V3 Tournament Edition 8K; Vietnam SMT → Vietnam. rulings.cbp.gov/ruling/N355069
  • N355986 (10 Dec 2025) — gaming controller; Thailand SMT → Thailand. rulings.cbp.gov/ruling/N355986
  • N356709 (17 Dec 2025) — Razer Kishi V3 / V3 Pro / V3 Pro XL; Thailand SMT → Thailand. rulings.cbp.gov/ruling/N356709
  • N359665 (25 Mar 2026) — Valve Steam Machine; motherboard made in Thailand imparts the character of the finished article → Thailand. rulings.cbp.gov/ruling/N359665
  • N360699 (29 Apr 2026) — Valve Steam Machine Bundle; origin of the set follows the console's essential character, retail packaging must mark each item. rulings.cbp.gov/ruling/N360699
  • N362101 (23 Jun 2026) — Razer Raiju V3 Pro; Thailand SMT → Thailand. rulings.cbp.gov/ruling/N362101
  • Full text of every ruling is searchable at the CBP CROSS database: rulings.cbp.gov
• U.S. International Trade Commission — Harmonized Tariff Schedule of the United States, Chapter 99 Subchapter III, headings 9903.05.77 (Thailand) and 9903.05.84 (Vietnam), rate column "the duty provided in the applicable subheading + 12.5%"; HTS 9504.50 general rate: Free. hts.usitc.gov

Note on method: the import and export series were pulled directly from the UN Comtrade public preview API and aggregated by partner country; the monthly figures reproduce to the cent on re-fetch. The mirror-statistics and neighbouring-code checks were run explicitly as attempts to falsify the headline, and are reported with their results. The origin findings and the rate-column wording are quoted verbatim from the CBP rulings and the HTS. The argument connecting the origin rulings to the import shift — that the customs record follows the board rather than the assembly line — is LostInConsoles' reading of those two public records, and is labelled as analysis. Forward-looking statements about the 12.5% surcharge are arithmetic on published rates and observed import values, not a projection of realised duty collections.

Excerpt: On 24 July 2026 a new U.S. Section 301 action — a forced-labor investigation covering 60 economies, published at 91 FR 47318 — began imposing an additional 12.5% on imports from Thailand (heading 9903.05.77) and Vietnam (heading 9903.05.84). It is not a "flat" rate: the tariff column reads "the duty provided in the applicable subheading + 12.5%." And the escape hatch — heading 9903.05.86, which exempts everything on Note 52's carve-out list — contains 9,692 tariff codes across 82 chapters and not one of them is in Chapter 95. The code list jumps straight from 9405.99.40 to 9620.00.50. Video game consoles (HTS 9504.50) are not exempt, not named, and not carved out. Meanwhile CBP has spent 2025 and 2026 issuing origin rulings that assign Razer, Turtle Beach and Valve controllers and consoles to Thailand or Vietnam — based on nothing more than where a circuit board was soldered. The industry migrated out of China to escape a tariff. A tariff with an entirely different rationale has now followed it there.

An original LostInConsoles data investigation, assembled from the USTR Section 301 notice at 91 FR 47318 and the Harmonized Tariff Schedule (Revision 19, 2026) Chapter 99 U.S. notes, cross-read against CBP CROSS country-of-origin rulings N349996, N350654, N355069, N355986, N356709, N362101, N347135, N359665 and N359663.

The short version

The console industry spent two years moving assembly out of China. The destination was narrow: northern Vietnam for Xbox and PlayStation, Thailand for components and for Valve's hardware. The legal instrument that made the move work was the CBP country-of-origin ruling — a document that decides, for tariff purposes, which country a product is "from."

Between April 2025 and June 2026, CBP issued a documented cluster of those rulings for exactly this hardware. Every one of them reached the same conclusion by the same route: the printed circuit board assembly (PCBA) is where the product's essential character lives, so the country that solders the components onto the bare board is the country of origin — even though the finished device is then shipped to China and assembled there.

Then, on 24 July 2026, a new tariff landed. It is not a China tariff, and it is not about semiconductors or reciprocal trade balances. It is a Section 301 action over the failure of 60 economies to enforce a prohibition on goods made with forced labour. Two of the named economies are Thailand and Vietnam. Both got the same rate: the ordinary duty plus 12.5%.

For HTS 9504.50 — video game consoles — the ordinary duty is Free. So the arithmetic is not complicated. The legal fiction that made a Thai-soldered controller "a product of Thailand" for tariff purposes now also makes it a product of Thailand for a tariff whose stated purpose has nothing to do with controllers at all.

And there is no carve-out. This is the part worth checking rather than assuming, because the notice's exemption architecture is real and extensive — it simply does not include the gaming shelf.

What the tariff actually says

The action was initiated on 12 March 2026 (60 investigations, published at 91 FR 12884) and concluded with a determination that 54 economies "have failed to impose and effectively enforce a prohibition on the importation of goods produced with forced labor," plus a second list of six economies that "have failed to effectively enforce" such a prohibition: Canada, Ecuador, the European Union, Indonesia, Mexico and Pakistan.

Thailand is economy 53 in the notice's determination. Vietnam is economy 60. Both are in the 54.

The rates are published as Chapter 99 headings, and the rate columns are identical in structure across the board:

Read that carefully, because "12.5%" is not a standalone duty and the difference matters. The heading does not replace the ordinary rate with 12.5%. It adds 12.5 percentage points to whatever the ordinary (MFN) rate already is. For a product whose base rate is 15%, the result is 27.5%. For a product whose base rate is Free — which is what HTS 9504.50 is — the result is 12.5%.

Every one of these headings carries the same lead-in clause:

"Except for products described in headings 9903.05.85–9903.05.92, articles the product of Thailand, as provided for in U.S. note 52 to this subchapter…"

That clause is the whole game. "Except for products described in headings 9903.05.85–9903.05.92" is the escape route. If a console is described in one of those headings, it escapes. So the only question that matters is what is in them.

The exemption list, counted

Heading 9903.05.86 is the general exemption. Its article description is blunt and self-referential: "Articles provided for in subdivision (b) of U.S. note 52 to this subchapter." So the exemption is not a principle. It is a list, and the list is in Note 52(b).

The operative sentence sits at the top of that subdivision:

"As provided in heading 9903.05.86, the duties imposed by headings 9903.05.20–9903.05.84 shall not apply to articles that are classifiable in the following provisions of the HTSUS…"

What follows is a mechanical enumeration of tariff codes. Run the census on the full list — not a window, not an excerpt, the complete subdivision from its opening to the end of the note — and this is what it contains:

• 9,692 code tokens
• 82 distinct chapters represented
• Chapters present at the chapter 92–98 boundary: 92 (0), 93 (0), 94 (44), 95 (0), 96 (3), 97 (19), 98 (6)
• Occurrences of the string "9504": 0
• Occurrences of "console": 0, of "game": 0, of "toy": 0

The list is not merely missing a video-game provision. It skips the entire chapter. Here is the verbatim sequence where the enumeration crosses that boundary:

...9405.99.20 9405.99.40 9620.00.50 9620.00.60 9802.00.40 9802.00.50 9802.00.60 9802.00.80 9818.00.05 9818.00.07

Chapter 94 ends at 9405.99.40. The next code is 9620.00.50 — chapter 96. Chapter 95, the chapter that contains 9504.50.00 (video game consoles and machines), does not appear at all, in any form, anywhere in the exemption list. Neither does 9503 (toys), nor 9504.20 (billiards), nor 9504.90, nor 9505 (festive articles), nor the 9506 sporting-goods block that sits beside them in every other Chapter 99 list.

This is worth pausing on, because Chapter 95 is not an obscure corner of the schedule. In the other Chapter 99 lists — the ones that govern the China Section 301 tariffs — Chapter 95 is enumerated exhaustively. The video-game line 9504.50.00 appears in the China list at two separate points in Chapter 99, where it is routed to heading 9903.88.16 under note 20(u)(i), the suspended List 4B tranche. Chapter 95 appears there in full: 9503.00.00, 9504.20.60, 9504.40.00, 9504.50.00, 9504.90.40, 9504.90.60, 9505.10.15 through 9505.90.60, the 9506 block, 9507.10.00 through 9507.90.80, and 9605.00.00. Someone drafting those lists knew exactly where the toys-and-games lines sit.

The forced-labor exemption list, compiled in 2026 by the same office, omits Chapter 95 entirely.

The list was not the product of inattention

If Note 52(b) were a rushed document, the chapter-95 gap could be read as an oversight. It is not a rushed document, and the surrounding sub-divisions show the drafting was deliberate and surgical.

Note 52's other carve-outs are narrow and specific, and each one is a category of goods someone thought through:

• Heading 9903.05.87 exempts a named list of particular articles — etrogs, frozen tropical fruit, and other individually-described goods.
• Heading 9903.05.88 exempts civil aircraft: airframes, engines, parts, components, sub-assemblies, and ground flight simulators.
• Heading 9903.05.89 exempts articles for use in pharmaceutical applications, by named tariff provision.
• Heading 9903.05.90 exempts aluminium, steel and copper articles and derivative steel/aluminium articles; passenger vehicles and light trucks, by heading.
• Heading 9903.05.85 is the in-transit transition: goods already loaded onto a vessel and in transit before 12:01 a.m. Eastern on 24 July 2026.
• Heading 9903.05.91 exempts humanitarian donations.
• Heading 9903.05.92 onward are the 54 individual country headings, beginning with Algeria.

Semiconductors get their own carve-out too, in Note 52 subdivision (f)(7), pointing to heading 9903.79.01. The Chapter 84 and 85 codes that do appear inside Note 52 belong to that semiconductor exclusion — they cover chips and chip-making inputs, not finished consumer electronics. There is no reading on which a game controller or a console is swept in by them.

So the machinery to exempt a category exists, is used freely, and is used precisely. Aircraft got a heading. Pharmaceuticals got a heading. Cars got a heading. Steel got a heading. A controller got nothing — not a heading, not a line, not a mention.

The rulings that made "made in China" stop being true

The reason this tariff lands on consoles at all, rather than on Chinese ones, is a body of customs law the games press has never read. It lives in CBP's CROSS database, and the cluster is unusually concentrated: one applicant, one product family, a sequence of rulings issued over fifteen months.

For Razer, the pattern is systematic enough to look like a compliance programme. Every product family got its own ruling, filed through Razer (Asia-Pacific) Pte. Ltd. in Singapore and signed by the same requester, Joe Castillo:

For Turtle Beach, ruling N349996 (1 Jul 2025) covers the Stealth Ultra Wireless XB/PC Global Controller, model TBS-0710-05 FG, with the dongle and charging dock — and finds the country of origin to be Vietnam.

For Valve, the rulings cover hardware the public has barely seen:

Valve's Steam Deck ruling is the most instructive, because it shows the split and still resolves it one way. The device's main PCBA is manufactured in Thailand, while three supporting PCBAs — input control boards, thumb stick boards, an audio board — are manufactured in China using SMT. CBP's conclusion:

"…the assembly of the main PCBA in Thailand by soldering the individual components onto the bare board results in a substantial transformation of the components to produce the main PCBA of Thai origin."

That is the whole mechanism in one sentence. It is not assembly of the finished device that confers origin. It is the soldering of components onto the board. Thailand holds the origin because Thailand did the surface-mount work, even though China built the boards that sit alongside the Thai one.

The formula, repeated word for word

Across the cluster, CBP repeats the same analytical formula. It is worth quoting in full because it is stated identically in every ruling, and it is explicitly about tariffs, not labelling:

"When determining the country of origin for purposes of applying current trade remedies under Section 301 and additional duties, the substantial transformation analysis is applicable. See, e.g., Headquarters Ruling Letter H301619, dated November 6, 2018. The test for determining whether a substantial transformation will occur is whether an article emerges from a process with a new name, character, or use different from that possessed by the article prior to processing. See Texas Instruments Inc. v. United States, 681 F.2d 778 (C.C.P.A. 1982). This determination is based on the totality of the evidence. See National Hand Tool Corp. v. United States, 16 C.I.T. 308 (1992), aff'd, 989 F.2d 1201 (Fed. Cir. 1993)."

The rulings do not merely determine origin as an abstract matter. They determine it for the purpose of applying Section 301 and additional duties — which is exactly the duty this new action imposes. The findings of fact they rely on are stated with a specificity that reads like a factory audit:

• N350654 (Razer Wolverine V3): "both the wired and wireless modes of the controllers contain a printed circuit board assembly (PCBA) that is manufactured in Vietnam via surface mount technology (SMT) process utilizing components which are sourced from various countries."
• N355986 (Kitsune): "the subject controller contains a printed circuit board assembly (PCBA) that is manufactured in Thailand via a surface mount technology (SMT) process."
• N362101 (Raiju V3 Pro): after the Thai PCBA is built, "the functional PCBA will be sent from Thailand to China. In China, the final assembly takes place, which consists of mounting the PCBA into a shell/casing, and installing other parts, such as brackets, buttons, directional pad, thumb sticks, connector, grips, battery, etc. Then, the completed game controllers are inspected, tested, labeled, and packaged."

And in each, the essential-character finding is phrased the same way:

"…it is the opinion of this office that the PCBA contributes to the main functionality of the finished merchandise."

For Valve's console it is stronger still. In N359665, CBP finds that

"the motherboard, which contains the essential functional components of the console, imparts the character of the finished article"

and adds the sentence that closes the door:

"Further, this PCBA is not substantially transformed into a new and different article of commerce by the processes performed in China."

Read those two rulings together and the manufacturing map of a modern console falls out. A device can have its motherboard soldered in Thailand, be shipped to China, have its chassis, thermal module and power supply assembled around it there, have its final PCBAs installed there, and be packaged for the United States there — and still, for U.S. tariff purposes, be a product of Thailand. The China assembly step is, in the law's eyes, invisible.

Why the migration and the tariff are now the same line item

Put the two documents side by side and the collision is not subtle.

The 2025–26 origin rulings exist because the industry needed its hardware to stop being a product of China. China's Section 301 tariffs sit on top of everything else in Chapter 99, and they are not refundable by moving a chassis. The CBP ruling is the receipt that proves the move worked: it certifies that the article is a product of Thailand or Vietnam, and therefore is not a product of China for the duties that punish Chinese origin.

Then the July 2026 forced-labour action made Thailand and Vietnam themselves tariffed origins — at the ordinary rate plus 12.5% — and exempted a list that omits Chapter 95 completely.

The result is a specific, arithmetic outcome for the products in the CBP cluster:

• A Razer Kitsune, Kishi V3 or Raiju V3 Pro — ruled a product of Thailand — is now a product of a +12.5% country.
• A Razer Wolverine V3 or Turtle Beach Stealth Ultra — ruled a product of Vietnam — is now a product of a +12.5% country.
• A Valve Steam Deck OLED, Steam Machine or Steam Frame — ruled a product of Thailand — is now a product of a +12.5% country.

And because HTS 9504.50 carries a Free general rate, there is no MFN margin to absorb it. Where a Chinese-origin console's duty burden is set by the China Section 301 layers, a Thai-origin console's burden was, until 24 July 2026, zero. The migration did not just reduce the tariff. It eliminated it. Now the same 12.5% applies to the destination that was chosen specifically to avoid a tariff.

There is one more clause that closes an escape route a customs lawyer would reach for first. Note 52(a) states that duties under these headings "shall also be subject to any additional duty provided for in this subchapter or in subchapter IV of chapter 99," and adds, explicitly, "Notwithstanding U.S. note 1 to this subchapter." The stacking is intentional. It also forecloses the classic out: articles entered under subheadings 9802.00.40, 9802.00.50, 9802.00.60 or heading 9802.00.80 — the outward-processing provisions used by companies that export a component, have it processed abroad, and re-import it — do not escape. The note says the additional duties "apply to the value of repairs, alterations or processing performed" under those subheadings, and to "the value of the article assembled abroad, less the cost or value of such products of the United States" under 9802.00.80. The outward-processing route is taxed, not exempted.

The part that is genuinely unresolved

Two things should be said plainly, because the temptation to over-read this is real.

First, the effective rate is not always 12.5%. The rate column adds 12.5 percentage points to the ordinary rate. For HTS 9504.50 that happens to be Free, so 12.5% is the effective figure for consoles. But a controller classified elsewhere, or a component with a non-zero general rate, faces more — and anyone reading "12.5%" as a ceiling rather than a surcharge has read the column wrong.

Second, the exemption architecture has one more layer that this analysis does not close. Note 52(a) also says the duties apply "Except as provided in headings 9903.05.85–9903.06.21 and in subdivisions (b) through (k) of this note." Subdivisions (b) through (k) is a large space, and only subdivision (b) — the 9,692-code list — has been counted here. Subdivision (f), for instance, is where the aluminium, steel, copper, passenger-vehicle and semiconductor exclusions live, and it is operative through its own headings. The decisive finding is narrower and firmer: the general exemption list in Note 52(b) contains no Chapter 95 codes at all, and the video-game line 9504 is not present anywhere in the USTR notice's own text — a scan of the notice returns zero occurrences of the string "9504". A category that is never named in the notice and never listed in the general exemption is a category that pays.

A third caveat belongs in the same paragraph. The rulings establish origin; they do not set the classification. A product ruled "of Thailand" still has to be entered under a heading, and the rate depends on that heading. What the cluster establishes beyond argument is the origin fact, and it is the origin fact — not the assembly geography the industry advertises — that the tariff is written against.

What to watch

The interesting question is no longer whether the industry moved production out of China. CBP's rulings settle that, product by product and model number by model number. The question is whether the destination survived the move.

Three things are worth tracking:

1. Whether new CBP rulings start reaching different conclusions. The cluster's logic is uniform: the PCBA confers origin. If a future ruling instead finds that final assembly in China does substantially transform a Thai-built board — or if CBP reverses on periodic verification, which every ruling reserves the right to do ("the material facts described in the foregoing ruling may be subject to periodic verification by CBP") — the origin flips back and the tariff profile changes with it.
2. Whether the exemption list is amended. The forced-labour action is an executable regime, not a statute, and its carve-out list was built in 2026. If a gaming-industry trade association files comments seeking a Chapter 95 line — as the steel, aircraft, pharmaceutical and vehicle sectors evidently did — the shape of the exemption is the shape of who lobbied.
3. Whether the next console platform gets a ruling at all. Razer's filings are the tell. Twelve months of rulings, one per product family, filed before launch, by an in-house applicant in Singapore, reads as a company that has decided origin certainty is a line item worth paying for. A platform holder that skips that step inherits whatever origin CBP decides its hardware has — after the entry, not before.

The migration to Thailand and Vietnam was sold to shareholders as supply-chain resilience. It was also, in the customs paperwork, a tariff-avoidance architecture — and the paperwork is unusually explicit about it, because the rulings say in their own text that they exist "for purposes of applying current trade remedies under Section 301 and additional duties." The tariff that arrived on 24 July 2026 has a different rationale, a different legal basis, and a different list of countries. It found the consoles anyway.

Sources

• Office of the United States Trade Representative — Notice of Actions in Section 301 Investigations of Acts, Policies, and Practices of Various Economies Related to the Failure of Each Economy To Impose and Effectively Enforce a Prohibition on the Importation of Goods Produced With Forced Labor — 91 FR 47318, document 2026-15181, published 28 July 2026 (Docket Nos. USTR–2026–0265, USTR–2026–0266). Additional rates applicable to entries on or after 12:01 a.m. Eastern, 24 July 2026: federalregister.gov · govinfo.gov (PDF)
• U.S. International Trade Commission — Harmonized Tariff Schedule of the United States, Revision 19 (2026), Chapter 99, Subchapter III: U.S. note 52 subdivisions (a) and (b); headings 9903.05.31 (China), 9903.05.77 (Thailand), 9903.05.84 (Vietnam), 9903.05.85–9903.05.92 (exemptions); note 20(u)(i) and heading 9903.88.16 (China List 4B). HTS 9504.50 general rate: Free. hts.usitc.gov
• U.S. Customs and Border Protection — CROSS rulings (country of origin):
  • N349996 (1 Jul 2025) — Turtle Beach Stealth Ultra Wireless XB/PC controller -> Vietnam
  • N350654 (18 Jul 2025) — Razer Wolverine V3 Pro / V3 Tournament Edition -> Vietnam
  • N352864 (15 Sep 2025) — Nintendo Switch game cartridges
  • N355069 (6 Nov 2025) — Razer Wolverine V3 Tournament Edition 8K -> Vietnam
  • N355986 (10 Dec 2025) — Razer Kitsune All-Button Arcade Controller -> Thailand
  • N356709 (17 Dec 2025) — Razer Kishi V3 / V3 Pro / V3 Pro XL -> Thailand
  • N362101 (23 Jun 2026) — Razer Raiju V3 Pro -> Thailand
  • N347135 (24 Apr 2025) — Valve Steam Deck OLED -> Thailand
  • N359665 (25 Mar 2026) — Valve Steam Machine -> Thailand
  • N359663 (6 Apr 2026) — Valve Steam Frame (model Deckard) -> Thailand
  • All available at the CBP CROSS database: rulings.cbp.gov

Note: the Chapter 99 notes, the exemption census and the rate-column wording were read from the Revision 19 (2026) HTSUS text; the rate structure "the applicable subheading + 12.5%" is quoted exactly from the heading columns. The origin findings are quoted from the CBP rulings themselves; the tariff-consequence analysis that connects them to the July 2026 action is LostInConsoles' reading, and is labelled as such. The 12.5% figure is a surcharge on the ordinary rate, not a standalone duty.

On 14 July 2026 the European Commission adopted a delegated act (C(2026) 5031) that rewrites Article 11(2) of the EU Batteries Regulation and hands eight categories of products a way out of the strictest consumer-repair mandate in the single market. The Commission's own summary says it added "six new categories" — wearables such as smartwatches, fitness trackers and smart glasses, electric toys, food-contact thermometer probes, ATEX equipment, on-body drug-delivery systems and roof-mounted machine telematics. A full-text scan of the adopted legal text returns "controller" 0, "game" 0, "console" 0, "gamepad" 0, "handheld" 0. Game controllers are not on the list, not in the recitals, and not carved out anywhere in the eight points. They are precisely the consumer-electronics category that got no relief — and Nintendo's own published spec sheet shows, in grams and milliamp-hours, what complying instead of being exempted costs.

The rule

Article 11 of Regulation (EU) 2023/1542 is the provision that made user-replaceable batteries a design requirement rather than a repairability preference. The operative text is blunt:

"Any natural or legal person that places on the market products incorporating portable batteries shall ensure that those batteries are readily removable and replaceable by the end-user at any time during the lifetime of the product."

The regulation defines "readily removable" down to the tool:

"A portable battery shall be considered readily removable by the end-user where it can be removed from a product with the use of commercially available tools, without requiring the use of specialised tools, unless provided free of charge with the product, proprietary tools, thermal energy, or solvents to disassemble the product."

Article 11 also requires that products "be accompanied with instructions and safety information on the use, removal and replacement of the batteries," published permanently on a public website.

This is not a future regime with a vague start date. Article 96(2)(a) of the same regulation states: "Article 11 shall apply from 18 February 2027." Article 95 repeals the old Batteries Directive (2006/66/EC) but keeps its Article 11 alive on a bridge "until 18 February 2027." The EU wrote the deadline into two separate articles so there would be no gap and no ambiguity.

A crucial scoping detail, and the one that explains everything the industry is doing: Article 11 binds the person who places products on the market. It is not a retrofitting rule. Hardware already sold before 18 February 2027 is untouched — a point Nintendo states for its own customers, and the reason the industry's response is a rolling, forward-going design change rather than a recall.

What the exemption actually exempted

Article 11(2) has always allowed a narrow list of products to make their batteries replaceable "only by independent professionals" instead of by the owner. The original regulation listed two: wet appliances and professional medical devices. The delegated act adopted on 14 July 2026 replaces that paragraph and expands it to eight points:

Six of the eight — (c) through (h) — are new, which reconciles the Commission's "six new categories" language with the act's eight points. And a safety gate is bolted onto the whole list:

"The derogations set out in points (a), (c), (d), (e), (f), (g) and (h) of this paragraph shall only be applicable where such derogation is required to ensure the safety of the user and the appliance."

The recitals are explicit about why each category earned an exemption, and the reasoning is instructive because it describes a category of products a controller is not. Wearables qualify because "the miniaturization of wearable devices and the portable batteries powering them may result in situations where the battery is so tightly encapsulated … that its removal may create a non-negligible risk of damage or piercing of the battery." Toys get a temporary pass because of Regulation (EU) 2025/2509's 2030 design rules. Food probes qualify because "leaked chemicals from the battery would contaminate food." ATEX equipment qualifies because removal "could result in a violation of the safety requirements." On-body delivery systems qualify because removal "could result in … a false administration of medication."

A game controller — a device the user grips, that contains a hand-sized battery behind a removable panel, and whose entire failure mode is a worn-out battery after years of vibration and charging cycles — fits none of these rationales. So the Commission did not name it.

Nintendo's spec sheet is the receipt

Nintendo is the only platform holder that has gone on the record with numbers. Its European support page — "Information about upcoming battery-related revisions to some Nintendo products" — begins:

"Starting summer 2026, in preparation for upcoming changes in European battery regulations coming into effect in mid-February 2027, selected Nintendo products in Europe will begin to be replaced on a rolling basis by revisions that contain a user-replaceable battery."

The page then publishes the design delta product by product. This is a manufacturer disclosing, in public, the physical tax that compliance imposes:

• Nintendo Switch 2 console — from 5,220mAh to 5,172mAh, "approximately 1% smaller." Weight from ~401g to ~411g. Its included Joy-Con 2 get user-replaceable batteries too.
• Joy-Con 2 pair — battery capacity unchanged, but each controller gains about 2g (~66g → ~68g, ~67g → ~69g).
• Nintendo Switch 2 Pro Controller — from 1,070mAh to 897mAh, "approximately 16% smaller." Weight drops ~7g.
• Nintendo 64 Controller for Nintendo Switch — capacity unchanged; ~1g heavier.
• Nintendo GameCube Controller for Nintendo Switch 2 — 500mAh → 525mAh, "approximately 5% larger"; 210g → 215g.

The Pro Controller is the number that matters. The controller with the most demanding power budget — wireless, with the longest runtime expectations — surrenders 16% of its battery capacity to make the cell a user-swappable part. Nintendo also commits that "Battery replacement kits for each product will be available to purchase from Nintendo Store in Europe in the future," which is the operational half of Article 11's requirement.

Nintendo says the quiet part plainly: "There is no difference in functionality between current products and revised products containing user-replaceable batteries." It is not a feature. It is a compliance tax paid in battery life.

The products Nintendo would rather kill than redesign

The same page does something more revealing than redesign the lineup. It publishes a discontinuation list. Under the heading "The below products will not be replaced by versions that contain user-replaceable batteries in Europe":

Nintendo Entertainment System (NES) Controller, Pokémon GO Plus+, Nintendo Switch, Nintendo Switch Lite, Nintendo Switch – OLED Model, Nintendo Switch Pro Controller, SEGA Mega Drive Control Pad, Super Nintendo Entertainment System (SNES) Controller.

And then: "Nintendo will no longer offer the above-named products on Nintendo Store after mid-February 2027."

The console-level confirmation is explicit: "From mid-February 2027, almost ten years after Nintendo Switch launched in March 2017, Nintendo will no longer sell to retailers hardware in the Nintendo Switch family of systems — specifically Nintendo Switch, Nintendo Switch Lite and Nintendo Switch – OLED Model. Sales of Nintendo Switch hardware on Nintendo Store will also end in mid-February 2027."

Read the two lists side by side and the shape of the regulation's real-world effect appears. Nintendo applied a make-or-buy calculation to its own catalogue. Where a product is still a going concern on the current generation — the Switch 2, its Joy-Con 2, its Pro Controller, the N64 and GameCube accessories — it redesigns and pays the battery penalty. Where a product is legacy or marginal — a nine-year-old console family, a €30 NES pad, a Pokémon GO Plus+ accessory — it withdraws the product from the EU rather than re-tool it. Article 11 does not just mandate swappable batteries. It is functioning as a product-line pruning mechanism for the European market, and it is the manufacturer, not the regulator, choosing what gets pruned.

There is a second-order effect worth naming: the Switch 1 family and its accessories are not being removed because they are unsafe or obsolete. They are being removed because the cost of making a 2017-era accessory EU-compliant exceeds its remaining EU revenue. The regulation produces the same visible outcome in Europe — a smaller catalogue — that a business decision would, and it does so through a safety-and-recycling measure that never mentions game consoles.

The door to an exemption is already shut

If a controller maker wanted the treatment wearables got, the procedural route is Article 11(4), which empowers the Commission to add products to the derogation list by delegated act. That route ran through a Commission Notice (C/2025/214, 8 January 2025) that opened a call for applications. The Commission's own batteries page states the window plainly and closes it: applicants "were invited to do so before 30 April 2025. The process is now closed."

That is the deadline that mattered. The delegated act adopted in July 2026 drew on "the evidence collected through a call for applications … open for three months between February and April 2025." If no controller application was filed and technically validated in that window, no controller appears in the act — which is exactly what the keyword scan shows. And the industry is now lobbying to get the door reopened: the European fire-and-security association Euralarm has publicly urged the Commission "to reopen the procedure foreseen in the 2025 Commission Notice to allow stakeholders to submit new requests for exemptions under Article 11(4)." Reopening it, after an act that spent its political capital on wearables and toys, would be a heavier lift than filing on time in 2025 was.

The Commission also flagged the constraint that will govern any future exemption request: "Small lithium-ion batteries not disposed of correctly are causing an increasing number of fires in waste treatment plants. This must be considered before granting exemptions for products containing such batteries." A game controller contains exactly a small lithium-ion battery. The exemption logic points away from controllers, not toward them.

The silence from the other two platforms

Sony and Microsoft sell controllers into the same 27-plus markets on the same 18 February 2027 clock, and the design problem lands differently on each:

• The DualSense uses a built-in, non-user-replaceable rechargeable pack. Under Article 11(1), that is the design Article 11 prohibits for new units placed on the EU market after the deadline.
• The Xbox Wireless Controller already runs on AA cells or the Play & Charge kit — user-swappable by construction — but Microsoft's own support documentation draws a line through the range: "the Xbox Elite Wireless Controller Series 2 and Xbox Adaptive Controller have built-in batteries."
• The Nintendo Switch 1 Pro Controller and Joy-Con, on the current design, hold built-in packs — which is why they are on the discontinuation list rather than the redesign list.

Despite repeated searches for a European redesign, a compliance statement, or a derogation application, no public Sony or Microsoft announcement of an Article 11 response has surfaced. That absence is itself the finding: the platform most exposed — the one whose flagship controller has the largest built-in pack and the longest claimed runtime — is the one that has said the most, and it said it by publishing a spec table and a kill list.

What to watch

The delegated act is not yet final law. It was adopted on 14 July 2026, transmitted to the European Parliament and Council for scrutiny, and — per the Commission — "will enter into force 20 days after its publication in the Official Journal of the EU, if the European Parliament or the Council fail to object to it." The act had already cleared an expert group in December 2025 and February 2026, public feedback from 28 April to 26 May 2026, and a WTO Technical Barriers to Trade notification from 12 June to 11 July 2026. Objection is unlikely but not impossible, and the two-month scrutiny window is the last moment at which a category could still be added or removed.

After that, the compliance clock runs to 18 February 2027 with no further relief. The interesting question is no longer whether consoles get an exemption — the text says they don't — but what the two silent platforms ship. A DualSense with a serviceable battery compartment, or a European SKU with a smaller cell, or a withdrawal, are the three options Article 11 leaves. Nintendo has already costed its choice. Sony and Microsoft have not told anyone theirs.

Sources

• European Commission (Environment) — "Commission adds exemptions to portable battery removal rules" (July 14, 2026): environment.ec.europa.eu
• European Commission (Environment) — Delegated regulation on the removability and replaceability of portable batteries (C(2026) 5031 final, adopted July 14, 2026): environment.ec.europa.eu/publications
• Official Journal of the EU / EUR-Lex — Regulation (EU) 2023/1542 concerning batteries and waste batteries (Articles 11, 95, 96; Strasbourg, July 12, 2023): eur-lex.europa.eu
• Nintendo of Europe — "Information about upcoming battery-related revisions to some Nintendo products" (support FAQ; Wayback snapshot September 15, 2026): web.archive.org
• Nintendo of Europe — same FAQ, live page: nintendo.com

Note: the delegated act C(2026) 5031 was adopted on 14 July 2026 and had not yet entered into force at the time of writing; it enters into force 20 days after publication in the Official Journal unless the European Parliament or Council object. Nintendo's FAQ is a reported manufacturer disclosure, not a confirmed regulatory contract.

Excerpt: On 11 September 2026 the EU Cyber Resilience Act's Article 14 switched on the strictest security-disclosure regime ever aimed at consumer hardware: a manufacturer must file an early warning on an exploited vulnerability within 24 hours, a detailed notification within 72 hours, and a final report within 14 days — or face fines up to €15 million or 2.5% of worldwide turnover. Consoles are squarely in scope. The regulation text contains the word "console" zero times, "video game" zero times, "gaming" zero times. And under Article 69(3) the obligation is retroactive: it lands on the entire installed base — every PS5, Switch and Xbox already sitting in European living rooms — not just new hardware. The console industry's disclosure culture was built on the opposite instinct, and Nintendo proved it one day before the clock started.

An original LostInConsoles investigation into the collision between the EU Cyber Resilience Act's new reporting clock and the way the console industry handles security — assembled from the regulation's own text (Regulation (EU) 2024/2847), ENISA's Single Reporting Platform, and the console security practices already on the public record.

The short version: For thirty years the console industry has treated security disclosure as something closer to a trade secret than a duty. Exploits arrive as jailbreak posts, homebrew exploits and quietly-shipped firmware updates; the interesting vulnerabilities are often found by the community and published only once a patch is out. On 11 September 2026, that culture acquired a legal deadline. Article 14 of the Cyber Resilience Act now requires manufacturers of "products with digital elements" sold in the EU to report an actively exploited vulnerability to a national CSIRT and to ENISA — within 24 hours for an early warning, 72 hours for the substance, and 14 days for a final report.

The regulation's definitions reach consoles without ever naming them. Article 2 applies the CRA to "products with digital elements made available on the market, the intended purpose or reasonably foreseeable use of which includes a direct or indirect logical or physical data connection to a device or network." A console that talks to the internet qualifies on its face. So does a controller with a wireless radio. The regulation does carve out medical devices, cars, aircraft, marine equipment, spare parts and defence products — and nothing else. Consoles are not on the exclusion list, and they are not named anywhere in the text.

Then Article 69(3) removes the last exit. The CRA's general application date is 11 December 2027, and Article 69(2) grandfathers products already placed on the market — but only from the general requirements, and only until they receive a "substantial modification." Article 69(3) then says the quiet part out loud: "By way of derogation from paragraph 2 of this Article, the obligations laid down in Article 14 shall apply to all products with digital elements that fall within the scope of this Regulation that have been placed on the market before 11 December 2027." Every console already in a European home is on the clock. Not from 2027. From last week.

What actually switched on

OBSERVED FACT: Article 71(2) of Regulation (EU) 2024/2847 sets the dates: "This Regulation shall apply from 11 December 2027. However, Article 14 shall apply from 11 September 2026 and Chapter IV (Articles 35 to 51) shall apply from 11 June 2026." The regulation was signed at Strasbourg on 23 October 2024 and entered into force on the twentieth day following publication.

OBSERVED FACT: Article 14(1) and (3) require a manufacturer to "notify any actively exploited vulnerability contained in the product with digital elements that it becomes aware of" — and any "severe incident having an impact on the security" of that product — "simultaneously to the CSIRT designated as coordinator … and to ENISA," via "the single reporting platform established pursuant to Article 16."

OBSERVED FACT: The deadlines are fixed and short. Under Article 14(2): "(a) an early warning notification … without undue delay and in any event within 24 hours of the manufacturer becoming aware of it"; "(b) … a vulnerability notification … within 72 hours"; "(c) … a final report, no later than 14 days after a corrective or mitigating measure is available." Severe incidents run on the same 24/72-hour pattern, with the final report due "within one month" (Article 14(4)(c)).

ANALYSIS: This is a reporting regime aimed at industrial software and connected-device makers, and it lands on consumer electronics with no adaptation. The 24-hour clock is not a target or a best practice — it starts when the manufacturer becomes aware, and "aware" is the moment an engineer reads the report. For a platform holder whose security posture has historically run through a bug-bounty scope document and a quarterly firmware cadence, the gap between "we know" and "we have told a government" just collapsed to a day.

The word "console" does not appear

OBSERVED FACT: A full-text search of the regulation (365,457 characters of the Official Journal legal text) returns the following counts: "console" = 0. "video game" = 0. "gaming" = 0. "game console" = 0. "handheld" = 0. The word "controller" appears 3 times, and every one of them refers to a data-protection controller under the GDPR or a microcontroller in the Annex — not a gamepad.

OBSERVED FACT: What the CRA covers instead is the generic category. Definition (1): "'product with digital elements' means a software or hardware product and its remote data processing solutions, including software or hardware components being placed on the market separately." Article 2(1) supplies the trigger — the "intended purpose or reasonably foreseeable use of which includes a direct or indirect logical or physical data connection to a device or network."

OBSERVED FACT: The exclusions are enumerated in Article 2 and do not include gaming hardware: medical devices (Regulation (EU) 2017/745 and 2017/746), motor vehicles (2019/2144), civil aviation (2018/1139), marine equipment (2014/90/EU), spare parts manufactured to the same specification, and products "developed or modified exclusively for national security or defence purposes." Article 2(5) allows the Commission to limit or exclude other sectors by delegated act where sectoral rules provide "the same or a higher level of protection" — but that requires an affirmative act, and no console-specific exclusion has been adopted.

ANALYSIS: This is the structural asymmetry at the centre of the story. The CRA reaches consoles through a definition broad enough to cover a smart doorbell, and the industry it catches is one that has never been regulated as a cybersecurity category at all. There is no "console" carve-out, no gaming annex, no sectoral regime waiting in the wings. The device in the living room is governed by the same text as an industrial router, and the text does not know what it is.

The retroactive clause is the whole fight

OBSERVED FACT: Article 69(2) creates the general transition: "Products with digital elements that have been placed on the market before 11 December 2027 shall be subject to the requirements set out in this Regulation only if, from that date, those products are subject to a substantial modification." In plain terms: old hardware is mostly grandfathered until you materially change it.

OBSERVED FACT: Article 69(3) then carves reporting out of that protection entirely: "By way of derogation from paragraph 2 of this Article, the obligations laid down in Article 14 shall apply to all products with digital elements that fall within the scope of this Regulation that have been placed on the market before 11 December 2027."

ANALYSIS: Read the two paragraphs together and the design becomes visible. The drafters grandfathered legacy hardware for the heavy conformity requirements — the technical documentation, the risk assessments, the CE-marking apparatus — because retrofitting those onto a ten-year-old product is impractical. But they deliberately refused to grandfather reporting, because a vulnerability in a product already in the field is exactly the hazard the article exists to catch. The result is that the CRA's reporting duty is the one obligation that reaches backward through the installed base. A PS5 bought at launch in 2020, a Switch bought in 2017, an Xbox Series X bought in 2020: all of it is in scope today. The regulation does not care when the box was sold, only that it is still supported — and Recital 60 expects it to be supported for a while yet.

OBSERVED FACT: Recital 60 states the support expectation: "The support period for which the manufacturer ensures the effective handling of vulnerabilities should be no less than five years, unless the lifetime of the product with digital elements is less than five years." Recital 57 adds that where a product "has a user interface or similar technical means allowing direct interaction with its users, the manufacturer should make use of such features to inform users that their product … has reached the end of the support period."

ANALYSIS: Five years is a floor, and consoles routinely exceed it — Nintendo's own support history runs into decades of hardware in the wild. The practical consequence is that the retroactive reporting duty and the multi-year support reality overlap on an installed base of tens of millions of devices per platform, each of which is now a reportable object.

The penalty is not symbolic

OBSERVED FACT: Article 64(2): "Non-compliance with the essential cybersecurity requirements set out in Annex I and the obligations set out in Articles 13 and 14 shall be subject to administrative fines of up to EUR 15 000 000 or, if the offender is an undertaking, up to 2,5 % of its total worldwide annual turnover for the preceding financial year, whichever is higher."

ANALYSIS: For a platform holder the percentage limb is the operative one, and it is measured against worldwide turnover, not EU revenue. Failing to file a 24-hour early warning is, on the face of the article, a finable infringement in the same bracket as failing the essential security requirements themselves. And the duty runs to "products … placed on the market" — the nexus is the EU market, not the location of the company. A Japanese or American manufacturer selling into Europe carries the obligation.

OBSERVED FACT: Article 14(7) explains where a non-EU manufacturer sends the notification. The default is the CSIRT of the member state where the manufacturer "has its main establishment in the Union," defined as where "the decisions related to the cybersecurity of its products … are predominantly taken." Where there is no main establishment, the article sets a cascade: the member state of the authorised representative acting for the most products, then the importer placing the most products on the market, then the distributor, then the member state "in which the highest number of users … are located."

ANALYSIS: That cascade answers the mechanical question and raises the strategic one. Nintendo, Sony and Microsoft all have European corporate arms — Nintendo of Europe GmbH in Frankfurt, Sony Interactive Entertainment Europe in London, Microsoft's Irish entity — so on paper the "main establishment" test resolves cleanly. But the test asks where cybersecurity decisions are predominantly taken, and for platform holders whose security engineering sits in Kyoto, Tokyo, Redmond and Foster City, that is a question each company now has to answer against a legal definition rather than a corporate-chart one. Article 14(7) does not let a manufacturer choose its regulator; it infers one.

Nintendo showed the shape of the problem one day before the clock started

OBSERVED FACT: On 10 September 2026 — the day before Article 14 went live — Nintendo published a security advisory for a high-severity flaw in the original Switch, tracked as CVE-2026-82079 and detailed publicly by security press on 14 September. The vulnerability is a stack-based buffer overflow in the console's local wireless networking code, affecting firmware earlier than 23.0.0. Nintendo assigned it a CVSS 4.0 base score of 7.0 and described an adjacent-network attack with low complexity, no privileges required and passive user interaction. Third-party databases listed an EPSS probability of roughly 0.16%. Exploitation required an attacker "within wireless range" who directly scans a QR code displayed on the console or TV, using the Album's "Send to Smartphone" function or the corresponding Mario Kart Live: Home Circuit workflow. The console maker said the flaw could not be used to extract information from the newer Switch 2, and that it was reported by external security researchers.

ANALYSIS: Two things about this matter. First, the trigger for the story is a researcher-reported flaw, not an observed attack — and under the CRA's own definition that distinction decides whether the 24-hour clock fires at all. Definition (42) is precise: "'actively exploited vulnerability' means a vulnerability for which there is reliable evidence that a malicious actor has exploited it in a system without permission of the system owner." An EPSS score of 0.16% is a forecast, not evidence of exploitation, and a researcher report is not an attack. So CVE-2026-82079, on the public record, sits just outside the Article 14 trigger.

ANALYSIS: Second, that is the point. The console industry's disclosure model produces exactly this artefact — a researcher finds a flaw, the vendor patches and publishes an advisory, the fix ships as a firmware update, and the whole thing is handled through normal coordinated-disclosure channels with the community. Nothing about that process needs a government. Article 14 changes the threshold the industry will have to operate against: the moment there is "reliable evidence" of exploitation, the same workflow now owes a national CSIRT and ENISA a filing within 24 hours, whether or not the vendor would have chosen to say anything publicly and whether or not a patch exists yet. Nintendo published a careful, well-scoped advisory on 10 September because that is what good coordinated disclosure looks like. One day later the same class of event, with one extra fact attached — evidence of exploitation — carries a fine.

The disclosure culture this lands on

The CRA is arriving at an industry whose security practices were shaped by an entirely different set of incentives. The evidence of that culture is already on the public record.

OBSERVED FACT: The lifecycle of a console generation is finite and the tail is quiet. Nintendo's 3DS and Wii U eShop closed in March 2023, and Nintendo ended security updates for those platforms in April 2024 — leaving a large installed base permanently outside the patch window. Microsoft closed the Xbox 360 store on 29 July 2024, the same year the platform turned nineteen. These are the products that Article 69(3) now reaches, and some of them have no vendor security pipeline left to run.

ANALYSIS: A retroactive reporting duty presupposes a company still maintaining the product. Where support has already ended, the installed base is in scope on paper but has no functioning compliance path — the manufacturer has to either extend support, or accept that a reportable vulnerability in an end-of-life console has nowhere to go. The CRA does not resolve this; it just makes the gap visible.

OBSERVED FACT: Console security research has historically run through community and bounty channels rather than regulatory ones — Nintendo maintains a HackerOne programme with an explicitly scoped list of accepted targets and exclusions, Sony publishes a vulnerability-disclosure policy and a security.txt contact, and the highest-profile console exploits (jailbreaks, homebrew loaders, firmware privilege escalations) reach the public through researcher write-ups and enthusiast forums as often as through vendor advisories. (These programme pages define what a researcher may test; the point here is the channel, not the contents.)

ANALYSIS: Scoped bounty programmes and community disclosure are not weaknesses — they are how the industry has done this well for years. But they are built on a private relationship between a vendor and a researcher, with the vendor controlling the timing and content of disclosure. Article 14 superimposes a statutory duty with a hard clock and a public-sector recipient on top of that relationship. The two can coexist, but only if the vendor's internal process can route a researcher's report into a 24-hour government filing the same day. Most console security operations were not built to do that.

Where the clock actually bites

OBSERVED FACT: Article 14(8) adds a duty that reaches past the regulator and straight to the player: "After becoming aware of an actively exploited vulnerability or a severe incident … the manufacturer shall inform the impacted users of the product with digital elements, and where appropriate all users, of that vulnerability or incident and, where necessary, of any risk mitigation and corrective measures that the users can deploy … where appropriate in a structured, machine-readable format that is easily automatically processable." Where the manufacturer fails to inform users "in a timely manner," the notified CSIRTs "may provide such information to the users."

ANALYSIS: This is the clause with consumer-facing teeth. A console vendor that quietly ships a silent firmware patch — the industry's default good-news story — now has a legal obligation to tell users about an actively exploited vulnerability, in machine-readable form, and to hand the CSIRT the authority to speak if it doesn't. The console business has spent two decades normalising silent updates as a feature ("your console was updated overnight"). Article 14(8) makes silence a compliance risk in the specific case where exploitation is confirmed.

OBSERVED FACT: ENISA's Single Reporting Platform — the Article 16 mechanism all of this runs through — was scheduled to reach operational status by 11 September 2026, the same date the reporting duty began. Reporting is by "electronic notification end-points" under Article 16(1), with the notification accessible simultaneously to ENISA.

ANALYSIS: The practical shape of the regime, then, is a single EU-wide intake that a platform holder's security team must feed within a day of confirming exploitation, with national CSIRTs as the coordinator and ENISA as the parallel recipient — and with the manufacturer obliged to inform users on its own initiative. It is a substantial piece of security infrastructure, and it was built for an industry that is only now learning it is inside it.

What this means for the console generation in your living room

The Cyber Resilience Act does not name a single console, handheld or gamepad. It does not have to. It defines a category — hardware with a data connection, sold in the EU — broad enough to capture the entire market, excludes everything it wants to exclude by explicit list, and then, in Article 69(3), refuses to let the reporting duty be grandfathered away with the ageing hardware.

The result is a regime that went live on 11 September 2026 and applies to consoles already sold, already owned, and — for the older ones — already past the end of their patch pipelines. The 3DS that stopped receiving security updates in 2024, the Wii U, the Xbox 360 whose store closed in 2024, the PS5 and Switch 2 on the shelf today: all of them fall within a text that lists medical devices, cars, planes and boats as its exclusions and never once considers whether a games console deserves a line of its own.

Nintendo published a careful vulnerability advisory on 10 September 2026, one day before the clock started, and it did so through exactly the coordinated-disclosure machinery the industry has always used. It was the right way to handle the flaw. Under Article 14, the next one — the one where someone can point to evidence of exploitation — is a government filing due inside twenty-four hours, a machine-readable notice to users, and a fine measured against worldwide turnover if either slips. The console industry has a security culture it built itself, over decades, on its own terms. As of last week, Europe has attached a deadline to it.

Primary source: Regulation (EU) 2024/2847 of the European Parliament and of the Council of 23 October 2024 on horizontal cybersecurity requirements for products with digital elements (Cyber Resilience Act), Official Journal of the European Union — full legal text. Definitions, scope and obligations quoted from Articles 2, 14, 16, 18, 64, 69, 71 and Recitals 40, 57, 60. Console ecosystem evidence: Nintendo security advisory for CVE-2026-82079 (10 September 2026), third-party security reporting on the advisory (14 September 2026), Nintendo/PlayStation/Microsoft vulnerability-disclosure programme pages, and publicly documented support-lifecycle end dates for 3DS/Wii U (eShop March 2023, security updates April 2024) and the Xbox 360 store (29 July 2024). ENISA Single Reporting Platform status per ENISA's published CRA reporting resources and contemporaneous security-industry coverage of the 11 September 2026 start date.

Excerpt: In January 2026 the White House imposed a 25% ad valorem duty on "semiconductor articles." The covered list is only three HTS headings — 8471.50, 8471.80, 8473.30 — and the operative subheading carves out "non-data center consumer electronics applications... including gaming." Consoles live at 9504.50, a line CBP has defended against the industry itself since 2002, when Sony argued the PlayStation 2 was a computer and lost. Meanwhile one year of U.S. customs import data shows the console supply chain that feeds that line collapse in China by 71% while Thailand — where Valve's own rulings say its console motherboards are made — grew 335%.

There is a tariff regime in effect right now that taxes computer parts at 25% and, by its own text, refuses to tax consoles at all. The reason isn't a loophole someone missed. It's a boundary that U.S. Customs and Border Protection has been drawing, and re-drawing, for a quarter century — and three documents published in the last eighteen months show exactly where that boundary sits.

The rule, in its own words

Proclamation 11002, signed January 14, 2026 and published at 91 FR 2441 (Federal Register Vol. 91, No. 12, January 20, 2026), imposed "an immediate 25 percent ad valorem duty rate on the import of certain advanced computing chips and certain derivative products described in the Annex to this proclamation (Covered Products)." Clause (3) of the operative text sets the rate and the effective moment: 12:01 a.m. eastern standard time on January 15, 2026.

The scope is the whole story. The proclamation defines "semiconductor articles" by cross-reference to U.S. Note 39, subdivision (b), to subchapter III of chapter 99 of the HTS. Extracting that subdivision from the government printing office PDF and searching the entire document for tariff headings returns exactly three: 8471.50 (processing units), 8471.80 (other units of automatic data-processing machines), and 8473.30 (parts and accessories of computers). The console headings do not appear anywhere in the proclamation. 9504 occurs zero times across the full text. So does 9503.

That is not an oversight in the drafting. The carve-outs are explicit and enumerated. Heading 9903.79.07, one of seven semiconductor subheadings in the new subchapter, reads:

"Semiconductor articles, as defined in subdivision (b) of U.S. note 39 to this subchapter, that are for use in non-data center consumer electronics applications in the United States."

And subdivision (e)(v) of the same note spells out what "non-data center consumer electronics" means: "applications... including gaming, personal computing, professional visualization, workstation applications, and automotive applications." Gaming is named. Consoles sit outside a tariff that was written, on its face, to reach the silicon inside computers. The rate table confirms the structure: 9903.79.03 covers data-center use, 9903.79.04 repairs, 9903.79.05 R&D, 9903.79.06 startups, 9903.79.07 consumer electronics including gaming, 9903.79.08 civil industrial.

The $6.7 billion line is 9504.50

So where do consoles actually enter? Heading 9504.50.00.00 of the HTS: "Video game consoles and machines, other than those of subheading 9504.30, and parts and accessories thereof." General rate of duty: Free.

That the console line is not the computer line is not new — and it is not uncontested. In HQ 964682, issued July 15, 2002, CBP's Commercial Rulings Division answered a request from Sony Electronics, filed on behalf of Sony Computer Entertainment of America, to classify the PlayStation 2 as an automatic data-processing machine under heading 8471. Sony's argument was architectural: the PS2's 128-bit Emotion Engine CPU ran at 300 MHz, executed ~6 billion floating-point operations per second, and carried its BIOS in a ROM chip on the main board — a computer by construction, argued the company that built it.

CBP disagreed, and said so in one word: "we do not agree." The PS2 was classified under 9504.10.00 — video games of a kind used with a television receiver — with the controller, AV multi cable and AC power cord classified alongside it, and even the plastic stands and multitap classified under 9504 by operation of Note 3 to chapter 95. The RF adapter alone escaped to 8543.89.9695. The holding has stood for twenty-four years, and it is the reason the device that made "console" a mass-market noun is not a "computer" for customs purposes — which is precisely the classification that keeps it out of the 2026 semiconductor tariff.

The boundary CBP drew against Sony in 2002 is the same boundary that decides, today, whether a device is a taxed "semiconductor article" or an untaxed consumer-electronics console. The industry has been trying to move that line ever since.

The line is moving again — this time for Valve

The freshest evidence that the boundary is live comes from three CBP rulings issued to Valve Corporation, all filed through customs counsel at Expeditors Tradewin.

N347135, dated April 24, 2025, concerns the Steam Deck OLED, described in the ruling as "a handheld gaming device" with a 7.4-inch HDR OLED touchscreen, a custom AMD APU, 16 GB of LPDDR5, Wi-Fi 6E, Bluetooth 5.3, a 45W charger, and SteamOS. The question was country of origin. Valve's process: the main printed circuit board assembly — the part the ruling calls "the central processing hub" responsible for "running the operating system, processing game applications, managing memory" — is manufactured via surface-mount technology in Thailand, then shipped to China, where three supporting PCBAs are made and the whole device is assembled into its housing with firmware loaded and final QC performed.

CBP held that the Thai board work is a substantial transformation and the Chinese assembly is not. The operative language: the main PCBA "is responsible for the core functionality of the gaming devices... Without the main PCBA, the device would be unable to perform its intended functions as a gaming console." The insert-and-screw work in China, CBP wrote, "is not complex and [is] considered basic assembly." Result: the Steam Deck OLED is a product of Thailand at import.

N360699, dated April 29, 2026 — barely four months ago — extends the same logic to the Steam Machine Bundle. Valve's own description of the product is instructive: "the Steam Machine Personal Computer (PC) gaming console," a cube that "is a high-performance PC that runs on a Linux-based operating system," which "can further act as a fully functioning PC as the user can install applications or alternative operating systems, browse the web, and expand the speed and functionality." Same manufacturing shape as the Steam Deck: main PCBA via SMT in Thailand, shipped to China, assembled into housing there, RAM and SSD sourced from Taiwan, Korea and Japan. Because the bundle puts a console and a controller of different origins in one box, CBP held the retail packaging "must indicate the proper country of origin for each item in the set" — "Console made in XXX" and "Controllers made in XXX."

Read together, the rulings describe a specific strategy: Valve keeps the value-creating step — the motherboard — in Thailand, and sends only the value-neutral step to China. The company that sells "a fully functioning PC" wants that PC's customs origin to be anywhere but the country the tariffs are aimed at.

What a year of customs data shows

The rulings are single-device snapshots. The aggregate picture comes from the U.N. Comtrade database, and it is dramatic. U.S. imports under HS 950450 — "video game consoles and machines" — pulled for full-year 2024 and full-year 2025:

Start with the top line: total console imports barely moved — $6.706B to $6.733B, up four-tenths of a percent. America did not stop buying consoles, and it did not stop importing them. What changed is where they came from.

China went from 86% of U.S. console import value in 2024 to 25% in 2025 — a collapse of $4.13 billion in a single year. Vietnam absorbed the bulk of it, rising more than eleven-fold. But the line that ties most directly to the CBP rulings above is Thailand: from $54 million to $236 million, up 335%. Thailand is where Valve's rulings place the Steam Deck and Steam Machine main boards. In unit terms the shift is even sharper: China shipped 47.0 million console units to the U.S. in 2024 and 21.3 million in 2025, while Vietnam went from 1.6 million units to 15.7 million.

Note the asymmetry between value and volume. China retained 21.3 million units in 2025 but only $1.657 billion of value — about $78 per unit. Vietnam shipped 15.7 million units for $3.754 billion, or about $239 per unit. The console categories moving to Vietnam are the high-value ones; what remains in China is lower-value.

Why this is the classification story, not just a tariff story

The temptation is to file this under "tariffs reshore supply chains," which is true but shallow. The deeper point is that the 2026 semiconductor action and the origin rulings are the same story told from two ends.

The tariff drew a line — 8471 computer articles on one side, taxed; consumer electronics including gaming on the other, carved out. The rulings drew a different but adjacent line — the motherboard's country on one side, assembly's country on the other. A device like the Steam Deck or Steam Machine sits on both lines at once: architecturally a computer, by CBP's naming a "gaming console," and by origin a Thai product assembled in China. Every one of those distinctions is worth real money, and Valve is paying counsel to fix them in writing.

The 2002 Sony ruling is the spine. Sony tried to have the PS2 declared a computer in a year when the tariff consequence was negligible. It failed. Two decades later the same classification question — is this thing a computer or a console? — sits directly on top of a 25% ad valorem duty and a multi-billion-dollar import flow. The companies that win are the ones whose devices land on the console side of a line CBP drew in 2002, or whose supply chain puts the value-creating step in a country the tariff isn't aimed at, or both.

What could move

The proclamation is explicitly a two-phase design. Phase one is the "very narrow category of semiconductors" now taxed at 25%. Phase two, to follow "after trade negotiations have concluded," is described in the text as "broader tariffs on semiconductors, at a rate of duty that is significant," paired with a "tariff offset program" letting companies that invest in U.S. semiconductor production "obtain preferential tariff treatment."

A narrow list of three headings is easy to keep narrow. A "significant" broader tariff with an investment-offset mechanism is not. The moment the covered list reaches beyond processing units and storage — toward complete articles, or toward the subassemblies and boards that sit inside them — the consoles that walked away clean in January 2026 are back in scope, and the 9504.50 vs. 8471.50 line stops being a quarter-century curiosity and becomes a line item worth the full $6.7 billion of annual U.S. console imports.

That is why the Thai motherboard in Valve's rulings matters more than it looks. It is a company pre-positioning its customs origin for a tariff that hasn't been written yet — using a boundary CBP first defended, against Sony, in 2002.

Method and sources

• Proclamation 11002 (91 FR 2441), full text including the Annex, extracted from the U.S. Government Publishing Office PDF; Note 39 subdivision (b) semiconductor-article list and the 9903.79.xx rate table read directly from the printed text. Searches for "9504" and "9503" across the complete document return zero matches.
• CBP rulings HQ 964682 (Sony PlayStation 2, July 15, 2002), N347135 (Steam Deck OLED, April 24, 2025), N360699 (Steam Machine Bundle, April 29, 2026), pulled from the CBP CROSS rulings database.
• HTS text for 9504.50.00.00, 8471.50.01, 8471.80 and 8473.30 pulled from the USITC HTS REST API.
• Trade flows from the U.N. Comtrade database, reporter United States (842), HS 6-digit 950450, imports (flow M), annual 2024 and 2025, by partner. Values are CIF; unit counts are the alternate-quantity field.

Excerpt: PLAION just pushed the NeoGeo AES+ back a full year to September 2027, blaming "unprecedented worldwide demand for memory chips, driven by investment in AI infrastructure." There's one problem with that explanation as a memory story: the AES+ is a hardware-accurate replica of a 1990 machine with 64KB of RAM. It buys less working memory than a smart lightbulb. And it's dying of the memory shortage anyway — because the scarce input isn't gigabytes, it's the obsolete little chips nobody wants to fab anymore. The same mechanism now reaches the cartridges themselves.

An original LostInConsoles investigation into why the AI memory crunch is now delaying and killing the smallest, least memory-hungry console hardware first — and why physical cartridges may be the next casualty.

The short version: For two years the story of the AI memory crunch in gaming has been a price story. DRAM triples, so systems get more expensive. That story is now too small. In September 2026 the crunch crossed into something nastier: it started cancelling products. And the counter-intuitive part — the part almost nobody has assembled — is which products are going first. Not the memory-hungry ones. The ones that use almost no memory at all. The NeoGeo AES+ carries 64KB of RAM and is delayed a year. A physical Evercade cartridge carries up to 1GB of flash and its maker says its entire business model is now about stockpiling "cartridge chips." The reason is that the memory market's victims are not defined by how much memory they buy, but by which node that memory is fabbed on.

The event: a 1990 console delayed a year, over memory

On 9 September 2026, PLAION REPLAI announced that the NEOGEO AES+ — a hardware-accurate recreation of SNK's 1990 AES, developed under license from SNK — would move its worldwide release to Thursday 16 September 2027. (Observed fact: the company's own press release.)

That is a full year of delay. And the reason PLAION gave is unambiguous. From the press release:

"Since announcing the NEOGEO AES+, PLAION REPLAI has been working with manufacturing partners to increase production capacity. However, unprecedented worldwide demand for memory chips, driven by investment in AI infrastructure, has created well documented industry-wide supply [constraints]… Taking additional time allows us to secure more memory chips while expanding our manufacturing."

The press release pairs memory scarcity with a manufacturing scale-up ("The decision to delay follows an extraordinary response to the announcement"). Pre-orders remain secure and the price is unchanged — this is a delay, not a price rise. The press framing that followed (Tom's Hardware, Shacknews, and others) collapsed this into "RAM shortage." Fine. But read that against the spec sheet and the whole thing gets strange.

The inversion: it barely has any memory

Here is the NeoGeo AES+ hardware, per the published specs:

• 64KB RAM
• 84KB VRAM
• 2KB audio RAM
• games read straight from ROM (the cartridge is the storage)
• the optional CD upgrade adds 7MB

Add it up: the entire working memory of the console is under 200 kilobytes. A 2026 midrange phone ships with 12 gigabytes of RAM — roughly 60,000 times more. The AES+ is not a memory-hungry device. It is nearly a memory-free device. Its design philosophy is the exact opposite of the thing that AI data centers are outbidding the world for.

So on the face of it, the vendor's explanation shouldn't hold. A machine that buys less memory than a thermostat cannot be the victim of a DRAM shortage — DRAM is sold by the terabyte and the AES+ buys by the kilobyte. This is the objection a sharp reader raises immediately, and it looks like it should kill the story.

It doesn't. It is the story.

The mechanism: it's not about gigabytes, it's about the node

The AI boom doesn't just make memory expensive. It makes specific kinds of memory extinct.

The memory that AI accelerators eat — HBM, high-layer 3D NAND, DDR5 — is fabbed on the same mature manufacturing lines that produce the cheap, low-density, legacy parts that small hardware runs on: SDRAM, SRAM, NOR flash, SLC NAND, DDR3/LPDDR3, eMMC. When a fab can sell a wafer of HBM to an AI buyer at several times the revenue of a wafer of legacy SDRAM, the economics decide the allocation. The legacy lines get converted or wound down. The obsolete little parts simply stop being made — not because demand for them collapsed, but because supply was reallocated to something more profitable.

This is documented as a structural shift, not a commodity blip. TrendForce's own research headline from June 2026 says it outright:

"Contract Prices Surged More Than 100% in 1H26; Structural Shortages to Keep NOR Flash and SLC NAND Prices Rising in 2H26."

That is the sentence that explains the NeoGeo AES+. NOR flash and SLC NAND are exactly the low-density, mature-node memory that a cartridge-based retro console lives on. Prices for those specific parts more than doubled in six months — and TrendForce calls the shortage structural, meaning it doesn't self-correct next quarter.

So the AES+ isn't being starved of "memory" in the abstract. It's being starved of cheap legacy memory on a node that the AI build-out is draining. It buys by the kilobyte — and kilobytes are precisely what's going out of production.

The second exposure: the cartridges are memory devices too

If the AES+ were just one delayed console, this would be a footnote. What makes it a pattern is the part the industry keeps missing: in a cartridge ecosystem, every single cartridge is itself a piece of memory hardware.

Consider Evercade. Blaze Entertainment's systems play games off physical flash-based cartridges that run 128MB to 1GB of flash each, sold for around $30, on hardware that costs $70–$200. In a July 2026 interview with FRVR, Blaze CEO Andrew Byatt described what the memory crunch has done to that model. He said the company had spent essentially all its cash hoarding components years in advance:

"We bought as much as we could in terms of cash flow… We bought as many cartridge chips, as much RAM, and as much processors and all the rest of it."

And the priority was explicit. Cartridges, he said, are "the lifeblood of the business" — so Blaze front-loaded cartridge chips for the next year and locked in 2027 supply, while treating console hardware as a "wait and see" item to buy as it goes.

That is a completely different shape of exposure from a normal console. An ordinary console maker has one bill of materials per box. A cartridge platform has two: the console BOM, and a media BOM for every cartridge in a growing catalog. When NOR/NAND flash prices more than double, the second bill gets hit not once but once per game, forever, across the whole back catalogue. And because Evercade deliberately keeps cartridge pricing consistent across 128MB and 1GB carts, the platform can't simply let its biggest-size carts float upward — the whole "buy a physical thing on a shelf" promise depends on that consistency.

Blaze has, for now, absorbed this by buying years ahead. It is a stockpile, not a solution. The stockpile is the story: a company chose to convert its cash into a warehouse of obsolete memory chips because the chips were the one input it could not count on getting later.

The pattern is bigger than one vendor

The AES+ and Evercade are the vivid cases, but they aren't isolated. In August 2026, the open-hardware maker PINE64 announced it was halting production of its Linux hardware line — boards, tablets, e-readers — blaming the DRAM/eMMC shortage directly on the AI build-out, expecting its existing PineNote and PineTab2 stock to run out within about three months and saying nothing new is coming before mid-2027. That's not a price rise. That's a product line going dark.

And the retro/niche segment has a whole casualty list now forming around the same cause: Retro Games Ltd's "The Spectrum" (delayed, with costs cited on chips and plastics), the Analogue Pocket (sitting at $239.99), Analogue's N64-based Analogue 3D, the Panic Playdate, Atari's 2600+/7800+ line, AYANEO's retro handhelds, and more. Some of these are primary-sourced, some are reported; treat the individual ones as provisional and the cluster as the signal. (Observed facts: the announcements. Inference, labeled: that they share one root cause — the vendors each name component/memory supply, not a single demand story.)

The pattern is consistent across every one of them, and it is the opposite of what a naive reading of "memory shortage" predicts: the smallest buyers of memory are the ones being cut off first. Not because they need a lot — because they need the wrong kind.

Falsification — the honest counter-cases

Objection 1 (the strong one): "A 64KB machine can't be a memory-shortage victim — memory is a cover story." We tested this hard, and it's the reason this article exists. The AES+ genuinely uses negligible working memory, and a raw commodity-DRAM shortage genuinely cannot explain its delay. But the objection only defeats a bad version of the story. The correct version isn't "AES+ ran out of RAM"; it's "AES+ is built on mature-node specialty memory, and that segment is being discontinued and repriced by the AI build-out — TrendForce's >100% NOR/SLC-NAND contract-price move is the receipt." The vendor named memory chips; the mechanism named by the data is the legacy node. So the objection doesn't falsify the thesis — it clarifies it. The inversion survives.

Objection 2: "It's really just demand — PLAION wants to make more consoles, and 'memory' is a euphemism." Partly fair: PLAION does say it is expanding manufacturing in the same breath. But you cannot build more consoles without the parts, and the vendor explicitly ties the delay to securing memory chips ("Taking additional time allows us to secure more memory chips while expanding our manufacturing"). Demand and component scarcity are complementary here, not competing. Treated as a co-factor; does not overturn the component cause.

Objection 3: "One delayed replica console isn't a market trend." Weakened by independent corroboration at a different product class (PINE64's boards/tablets on DRAM/eMMC) and a different exposure (Evercade's cartridge chips). Three unrelated vendors, three different memory types, one direction. The trend claim rests on that spread, not on the AES+ alone. Acknowledged limitation: not every retro casualty is primary-verified, and some price moves carry tariff/FX noise.

What the data says it means (analysis, not fact)

• Observed facts (high confidence): the AES+ delay to 16 Sep 2027 and PLAION's stated cause; the AES+ spec (64KB/84KB/2KB, ROM-fed); TrendForce's >100% 1H26 NOR/NAND contract-price rise and "structural" framing; Evercade carts at 128MB–1GB/~$30 and Byatt's "cartridge chips… lifeblood" quote; PINE64's halt on DRAM/eMMC through mid-2027.
• Inference (moderate, labeled): the binding constraint for retro/niche hardware is mature-node specialty memory, not commodity DRAM; cartridge platforms face double exposure (console BOM + per-cartridge flash BOM); the smallest memory buyers are structurally first in line to be cut.
• Speculative (explicit): that this becomes an extinction event for the smallest retro makers; that the cartridge-revival era is ending. TrendForce sees the shortage as structural into 2027 and beyond, which makes some attrition likely — but how far it goes is not knowable from here, and we flag it as such.

Bottom line

The NeoGeo AES+ is a 1990 console rebuilt in 2026 that carries less memory than a disposable gadget — and it just got delayed a year because the world doesn't make enough obsolete memory anymore. That sounds like a paradox. It's actually the cleanest evidence yet of what the AI memory crunch really is.

It isn't a shortage of memory. It's a shortage of the specific, unfashionable, low-margin memory that small hardware is built on — the SRAM, the NOR flash, the SLC NAND, the legacy SDRAM that AI has no use for and fabs no longer want to make. And because cartridges are memory, the crunch reaches the retro market twice: once when you can't build the console, and again when you can't afford to build the games.

The consoles that use the least memory are going first. That is the inversion, and it is where the story goes next.

Sources

• PLAION REPLAI — "The NEOGEO AES+ worldwide release date moves to 16th September 2027" (September 9, 2026): presse.plaion.com
• TrendForce — "Contract Prices Surged More Than 100% in 1H26; Structural Shortages to Keep NOR Flash and SLC NAND Prices Rising in 2H26" (June 16, 2026): trendforce.com
• Kotaku — "RAMageddon Is Hammering Retro Handheld Makers Like Evercade": kotaku.com
• Tom's Hardware — "Hardware-accurate NeoGeo AES+ delayed to late 2027": tomshardware.com
• GamesRadar — "The Neo Geo AES+ now isn't arriving until September 2027": gamesradar.com
• Shacknews — "NEOGEO AES+ delayed to 2027": shacknews.com
• IT'S FOSS — "Pine64 halts Linux hardware manufacturing over DRAM/eMMC prices" (August 2026): itsfoss.com
• Gaming.net — "NeoGeo AES+ delayed to September 2027 as PLAION cites memory chip supply": gaming.net
• Evercade / Blaze Entertainment — cartridge format and memory sizes, official product information: evercade.co.uk
• Wikipedia — Neo Geo (hardware specification: 64KB RAM / 84KB VRAM / 2KB audio RAM): en.wikipedia.org

Evercade's cartridge-flash sizes (128MB–1GB) and CEO Andrew Byatt's "cartridge chips… the lifeblood of the business" quote are as reported by Kotaku, drawing on the Blaze/FRVR interview. TrendForce's figures describe a reported industry-wide contract-price move, not a confirmed contract.

LostInConsoles is an independent investigations outlet focused on the global console hardware business. Quotes and dates are drawn from vendors' own releases and named interviews; causal readings that go beyond what sources state are labeled as analysis. Prices in USD unless noted.

Excerpt: In 2025 Nintendo, Sony and Microsoft raised American console prices and let the tariff story do the explaining. On February 20, 2026 the Supreme Court ruled those IEEPA tariffs unlawful, 6-3, and by August more than $100 billion had been refunded to importers. Nintendo booked roughly $300 million of it as a reduction in cost of sales; Sony expects about $508 million, most of it to its games business. No U.S. console price came back down. Instead Nintendo ran a 30%-off "Customer Appreciation Sale" funded "in part by tariff-related refunds," and in a July 2026 filing told a court that customers got "exactly what they bargained and paid for" and have no legal right to a rebate — while its lawyers disputed that tariffs were responsible for the specific price increases at all. Those two positions cannot both be true, and the contradiction is sitting inside the companies' own paperwork.

An original LostInConsoles data investigation into the tariff-refund windfall that U.S. console buyers never saw, assembled from earnings disclosures, the Learning Resources v. Trump docket, the Nintendo and Microsoft motions to dismiss, and the price-increase statements the industry issued in 2025.

The short version: The console industry's 2025 U.S. price increases were sold to the public as tariff arithmetic. Sony said it was "navigating a challenging economic environment" and raised the PS5 by $50 on August 20, 2025. Microsoft took the Xbox Series X from $499.99 to $599.99 in May 2025 and again to $649.99 in October. Nintendo nudged the original Switch to $339.99 and let Switch 2 accessories climb. Then the legal foundation under those increases was deleted: on February 20, 2026, the Supreme Court held 6-3 in Learning Resources, Inc. v. Trump that the International Emergency Economic Powers Act does not authorize the President to impose tariffs. Every IEEPA duty collected since 2025 became, retroactively, money the government was never owed. The Court of International Trade ordered Customs and Border Protection to return roughly $165 billion; by August, more than $100 billion had gone out.

The refunds landed as income-statement credits, not price cuts. Nintendo recorded approximately $300 million and told investors the refunded tariffs were "primarily borne by Nintendo rather than passed through in product prices" — a sentence that is, on its face, a defense of the price hikes. Sony told a court it expects about $508 million, ~70% already collected. No SKU came down. In September 2026 Nintendo converted part of the windfall into a two-week Customer Appreciation Sale (Sept 13–26) of 30%-off software, DLC, accessories and amiibo — a demand-stimulus move into its high-margin digital business, not a hardware rebate. And in the class-action defending that decision, Nintendo's filing argues buyers "received exactly what they bargained and paid for," a position that only works if the tariffs were never built into the price. But the tariffs were the stated reason for the price.

The refund is real, large, and visible in the accounts

OBSERVED FACT: Nintendo's August 2026 first-quarter investor presentation recorded approximately $300 million of tariff-related refunds tied to duties imposed under IEEPA, booked as a reduction in cost of sales. The same materials stated that the refunded tariffs were "primarily borne by Nintendo rather than passed through in product prices." The refund helped first-quarter operating profit rise 150.5% year over year to ¥142.5 billion even as net sales fell 9.5% to ¥517.8 billion (Nintendo IR; secondary coverage: ts2.tech).

OBSERVED FACT: Sony expects approximately $508 million in U.S. tariff refunds, the majority benefiting its Game & Network Services segment; about 70% had already been collected at the time of reporting (WinCentral summary of the Sony litigation).

ANALYSIS: A 150.5% operating-profit jump on falling revenue is the signature of a one-time credit, not of a healthy quarter. That is exactly how a windfall behaves: it flatters the margin, then it is gone. The refund was never a recurring revenue stream, and Nintendo's own framing — "primarily borne by Nintendo" — concedes it retained the benefit rather than routing it to the people who paid the higher price.

The Supreme Court deleted the justification, not the price

OBSERVED FACT: The Supreme Court decided Learning Resources, Inc. v. Trump on February 20, 2026, holding 6-3 that IEEPA does not authorize the President to impose tariffs. The Court of International Trade then ordered CBP to return roughly $165 billion; the Penn Wharton Budget Model put the ceiling at $175 billion; by August 2026 more than $100 billion had actually been disbursed (Forbes/Business Model Analyst accounting of the refund wave).

OBSERVED FACT: The price increases were justified to consumers in tariff-adjacent language. On August 20, 2025 Sony raised U.S. PS5 prices by $50: Digital Edition $450 → $500, disc model $500 → $550, PS5 Pro $700 → $750. Sony's statement: "Similar to many global businesses, we continue to navigate a challenging economic environment" — it pointedly did not name tariffs (CNBC, Aug 20 2025; The Verge).

OBSERVED FACT: Microsoft moved the Xbox Series X to $599.99 (from $499.99) in May 2025, then to $649.99 in October 2025 (adrenaline; lesnumeriques; Kotaku).

ANALYSIS: When the court removed the legal basis for the tariffs, it did not touch the prices. The increase had already been absorbed into the market as the new normal, and the refund became a separate line item flowing back to the seller.

Nintendo's own filing says the tariffs never raised the price

This is the hinge of the story, and it is a contradiction inside the companies' paperwork.

OBSERVED FACT: In a filing dated July 20, 2026, Nintendo moved to dismiss a U.S. class action seeking tariff rebates, arguing customers "received exactly what they bargained and paid for" when they purchased at the higher prices (Nintendo Life, Jul 2026; Gaming.net; GamesRadar — sourced to Game File).

OBSERVED FACT: Sony and Microsoft, in separate U.S. suits, argue they are not legally required to return tariff-related money to customers, and challenge the claim that the tariffs were actually responsible for the specific console price increases. Sony's motion argues that paying the market price for a voluntarily purchased product does not, by itself, constitute a legally recognizable injury (WinCentral).

OBSERVED FACT: The Microsoft case is Hastings v. Microsoft, brought by Xbox purchaser Trevor Hastings of Chula Vista, California, filed in King County Superior Court and removed to federal court in July 2026; it alleges Microsoft passed tariff costs to customers while standing to gain from the federal refunds (CPA Practice Advisor / Seattle Times).

INFERENCE — the two claims cannot coexist:

• Claim A (the 2025 marketing): prices went up because tariffs raised our costs.
• Claim B (the 2026 defense): the tariffs did not drive the specific price increases, and customers got fair value, so no rebate is owed.

If A is true, the refund restored a cost the buyer was made to cover, and keeping it is a windfall taken from consumers. If B is true, then the public tariff justification for the increases was never accurate. One of the two statements misled someone. The companies are not required to resolve it — but the record now contains both.

The refund went into a sale, not into the price

OBSERVED FACT: Nintendo of America announced a Customer Appreciation Sale running September 13–26, 2026 — "one of our largest promotions ever" — offering 30% off dozens of digital games and bundles, plus savings on DLC, physical games, accessories and amiibo. The press release states the sale "is made possible in part by tariff-related refunds. While Nintendo absorbed most tariff-related costs, the refunds helped make promotions like this one possible" (Nintendo Life; ts2.tech).

ANALYSIS: This is the cleanest tell. The refund was recycled into software, DLC and accessory discounts — the highest-margin, most lock-in-heavy part of Nintendo's catalog — not into the console price that carried the increase. It is demand stimulus dressed as gratitude. A buyer who paid $50 more for hardware in 2025 receives, in 2026, a two-week discount on digital goods they might not have bought otherwise. The refund's path runs toward more spending, not toward restitution.

OBSERVED FACT — the contrast case: Amazon booked a $600 million tariff refund and told investors it would return part of it directly to customers, automatically, with no claim form — CFO Brian Olsavsky said on the Q2 earnings call that Amazon had "largely absorbed" the costs itself but would contact affected buyers in a "limited set of circumstances" and issue automatic refunds (IBTimes UK). Apple recorded close to $2.2 billion and said refunds added $0.11 to diluted EPS; General Motors took $500 million; Nike told investors in June it expected close to $1 billion; Kohl's reported gross margin up 305 basis points, of which only about 5 basis points were real once refunds are excluded (Business Model Analyst).

ANALYSIS: Amazon shows the pass-through is operationally possible — the same government event, handled by refunding the buyer. The console makers chose the other path: book the credit, defend the price, run a sale.

Falsification: does the memory crunch explain why prices never fell?

The strongest honest objection deserves to be stated, not buried.

TEST — "The refund is smaller than the new cost wave, so a price cut was never possible." This is partly right, and it is the real reason nobody cut prices. The refunds are one-time and modest; the cost pressure that replaced them is structural. Nintendo has flagged roughly $700 million in AI-driven memory cost inflation, and analysts broadly expect consoles not to get cheaper — with Switch 2 possibly getting more expensive — because DRAM/component costs are still climbing. Verdict: PARTIAL — it does not falsify the core claim. It explains why a price cut was unrealistic; it does not explain why the refund was framed as "primarily borne by Nintendo" while buyers who were told tariffs caused the increase receive nothing. The objection changes the remedy, not the contradiction. If anything it sharpens it: the refund that could have offset the memory wave for consumers was instead absorbed into margin and a promotional calendar.

TEST — "The companies never actually said tariffs caused the price." Sony's August 2025 statement avoided the word "tariffs," using "challenging economic environment" instead. Verdict: SURVIVES as a nuance, weakens nothing. The timing — a $50 increase the week tariffs took effect on most countries, after Sony said in May 2025 it was considering hikes to cover the administration's tariffs — is the substance; careful wording is the cover. And Nintendo's own investor language ties the refund directly to tariff-bearing, which only matters if tariffs were in the cost base.

TEST — "Nobody is legally owed anything, so there is no story." Possibly true as law. Verdict: does not falsify. The story is not that a court will order refunds; it is that the industry's public and legal positions on tariffs are mutually exclusive, and the money that flowed back was kept. That is an accountability gap, and it is documented on the record.

Why this matters beyond one sale

ANALYSIS — the time horizon. By the summer of 2027, importers must lap a refund they will not receive again for a tax some of them are still paying under replacement authorities, including Section 232 semiconductor tariffs. A one-time credit compares badly against a recurring cost. That is the setup for the next round of price increases — justified, this time, by memory and semiconductors rather than IEEPA — with the 2025 refund already spent and the 2025 increase already permanent. The console buyer absorbed the increase when it was convenient to blame tariffs, and will absorb the next one when it is convenient to blame AI memory. The refund never made it back to the receipt.

The bottom line: Nintendo and Sony booked an estimated ~$800 million in tariff refunds between them; no U.S. console buyer saw a dollar of it. The companies argue in court that the tariffs never caused the specific price increases they publicly blamed on tariff-era conditions — which, taken seriously, means either the refund belongs to consumers or the justification never did. Nintendo answered the question it preferred to answer, with a two-week sale on digital games and a filing that says you already got what you paid for.

Sources are linked inline and labeled OBSERVED FACT, INFERENCE, or ANALYSIS. Financial figures are the companies' own disclosures as reported and should be treated as reported values; the Microsoft refund total was not pinned to a primary figure at publication. The Learning Resources v. Trump date and holding, the refund volumes, the price points, and the court-filing language were verified against the sources cited.

Excerpt: When Nintendo finally switched on docked VRR on the Switch 2 in firmware 23.0.0 (September 9, 2026) — roughly 16 months after the console launched at $449.99 — journalists reported it as overdue. The more revealing fact sat in the teardown ledger all along: the Switch 2 dock uses a Realtek RTD2175N DisplayPort-1.4-to-HDMI-2.1 protocol converter, a member of the same Realtek RTD2175 display-bridge family that Sony ships inside the PlayStation 5. Both consoles therefore launched with VRR-capable silicon in the video path and both delivered the feature years late as a firmware unlock — Sony about 17 months after the PS5's November 2020 launch, Nintendo about 16 months after the Switch 2's June 2025 launch. No major outlet has assembled this cross-console silicon pattern, and it changes how the Switch 2's rollout should be read: not a Nintendo-exclusive failure, but a recurring flaw in a shared third-party part — made worse by Nintendo's decision to make docked VRR work on a per-game whitelist that, at launch, exactly one app supports.

An original LostInConsoles data investigation into the shared Realtek display-bridge silicon inside the PS5 and Switch 2, the firmware-gated VRR history on both consoles, and the per-game whitelist that defers the feature to an indefinite patch backlog.

The short version: The Switch 2's docked VRR rollout on September 9, 2026 was treated as a long-delayed feature finally arriving. But the component manifest says it was always a firmware decision, not a hardware one. The dock's Realtek RTD2175N (a DisplayPort 1.4 → HDMI 2.1 protocol converter) is from the same RTD2175 family as the RTD2175P HDMI encoder Sony uses in the PS5. The Verge proved in July 2025 that the RTD2175N in the dock could already carry a VRR signal — by plugging a Steam Deck into an official Switch 2 dock and watching VRR engage. Sony's PS5 went through the identical playbook: it shipped in November 2020, its HDMI path (Realtek RTD2175-family silicon) supported VRR, and Sony withheld the feature behind a system update that arrived in April 2022 — about 17 months later. Nintendo then repeated it. The difference is the implementation: Sony's PS5 VRR works system-wide (including an "apply to unsupported games" toggle), while Nintendo's docked VRR runs on a whitelist that requires every game to be individually flagged — so on day one, only the bundled Welcome Tour app (v1.1.0, its VRR slider demo) actually uses it.

The silicon connection that no outlet has assembled

The RTD2175N in the Switch 2 dock and the RTD2175P in the PS5 are not the same chip — repair-and-parts sources (which handle failures of these parts daily) are explicit that the RTD2175N is a DisplayPort-to-HDMI conversion chip in the Switch 2 dock while the RTD2175P is an HDMI encoder in the PS5/PS5 Slim mainboard, and that they are not interchangeable (kasynparts). But they are the same RTD2175 base family from Realtek, a third-party display bridge vendor whose parts sit in the video output path of the two best-selling consoles in the current generation.

This is precisely the kind of shared-supplier detail that is scattered across enthusiast teardown threads and repair forums rather than assembled into one data story. ResetEra's teardown thread identifies the Switch 2 dock part as the RTD2175N-CG; gbatemp reports the PS5 Slim's EDM-044 board uses the RTD2175P; a Famiboards thread on docked VRR puts it bluntly: "The PS5 uses the same RealTek chip (RTD2175) that does Display Port to HDMI function. The Sony one is RTD2175P and Nintendo's is RTD2175N. It was exact same issue with the Display Port > HDMI thing fucking up VRR on PS5."

OBSERVED FACT: Both the Switch 2 dock (RTD2175N) and PS5/PS5 Slim (RTD2175P) carry Realtek RTD2175-family display-bridge parts in their HDMI output path (source: ResetEra teardown thread, gbatemp, NeoGAF, parts vendors). INFERENCE: The VRR gating on both consoles was a firmware/software decision layered on top of silicon that already supported the HDMI 2.1 VRR signaling — not a hardware shortage that had to be engineered away. SPECULATION: Neither Sony nor Nintendo commented on whether the shared Realtek part contributed to the rollout delays; this is the assembly we are proposing, not a claim either company has acknowledged.

The 16-month gap, step by step (time-horizon reconstruction)

Reconstructing the timeline from the record:

• T-5yr → T-2yr (2021–2023): Sony's PS5, launched November 2020, is confirmed to have VRR-capable HDMI 2.1 silicon. Sony states VRR "will be added" via a future update; it actually lands in April 2022 (22.01-05.50), ~17 months after launch, alongside an "apply to unsupported games" toggle. (Push Square, GameSpot)
• 2024: Switch 2 is in late development; component manifests of the dock surface the RTD2175N-CG, confirming a Realtek RTD2175-family DP→HDMI 2.1 bridge capable of VRR/Adaptive-Sync, DSC and up to 4K120. (ResetEra)
• T-1yr (2025): Nintendo's US website advertises VRR for both handheld and docked modes; the docked wording is removed from the site; Nintendo apologizes, calling the docked-VRR mention "the incorrect information", and tells press the feature is "handheld mode only." (Nintendo Life)
• T-6mo (July 2025): The Verge's Sean Hollister proves the dock's hardware already carries VRR by attaching a Steam Deck to the official Switch 2 dock and detecting a variable-refresh signal — while the Switch 2 itself is still VRR-disabled when docked. (The Verge)
• T-30d→now (Sept 9, 2026): Firmware 23.0.0 enables docked VRR, requiring a separate dock firmware update (Settings > System > Update Dock), and activating it only for games on a whitelist. At launch, only the preinstalled Switch 2 Welcome Tour (v1.1.0, VRR slider demo) is whitelisted. (Nintendo Life, Digital Foundry)

OBSERVED FACT: The same physical dock units sold from June 2025 acquire VRR purely through a dock-firmware update — no dock hardware revision is required (source: Nintendo Life patch notes instructing a dock update, Digital Foundry's confirmed working VRR via the existing dock). OBSERVED FACT: Docked VRR is gated behind a per-game whitelist; at launch only Welcome Tour is whitelisted, per Digital Foundry's John Linneman (source: WCCFTech quoting Linneman, Digital Foundry). INFERENCE: Because the July 2025 Steam Deck test proved the launch dock could already carry VRR, the ~16-month gap is attributable to Nintendo's software/certification readiness and its chosen gating policy, not to missing silicon.

Why the whitelist is the real story, and where it departs from Sony

The technical community quickly noticed one flattering difference: in 60 Hz mode the Switch 2's docked VRR window runs 40–60 Hz — matching Xbox and beating the PS5's 48–60 Hz floor — with full LFC in 120 Hz mode. WCCFTech and Digital Foundry both flagged that Nintendo's range "soundly beat the PlayStation 5 implementation." But that win comes with a structural catch nobody framed as the counterweight:

• System-wide vs. per-game. The PS5's VRR, once shipped, is a system feature (with an "apply to unsupported games" toggle). The Switch 2's docked VRR is a whitelist: each title must ship a flag/patch declaring VRR support before the console will vary its refresh rate for that title.
• Day-one whitelist = exactly one app. At launch the only whitelisted software is the bundled Welcome Tour demo. Tweaktown, Digital Foundry and the community all confirm games "need to be patched" and that adoption "will take a while."
• The economics invert the benefit. The feature is most valuable to already-shipped, framerate-unstable titles in the Switch 2's catalogue — but those are precisely the titles whose publishers must now spend QA/certification time and per-title patch cycles to flag VRR, with no new revenue tied to doing so. The Welcome Tour bundled app is the only title with an incentive to be ready on day one.
• FreeSync owners are excluded. Community reports and ResetEra confirm the Switch 2's docked VRR is HDMI 2.1 VRR only — monitor owners relying on FreeSync over HDMI cannot enable it at all, a gap the system-wide-toggled rivals do not share in the same way.

OBSERVED FACT: Docked VRR operates on a whitelist requiring per-game developer flags/patches; only Welcome Tour (v1.1.0) is whitelisted at launch (source: Digital Foundry, WCCFTech, Tweaktown, tech-insider). OBSERVED FACT: VRR range is 40–60 Hz in 60-Hz mode with full LFC at 120 Hz, better than the PS5's 48–60 Hz floor (source: Digital Foundry, Nintendo Life comments). OBSERVED FACT: Docked VRR is HDMI 2.1 VRR only, not FreeSync (source: ResetEra). INFERENCE: Nintendo's per-game whitelist is a deliberate compatibility-safety mechanism (avoid forcing VRR on titles with inconsistent frame pacing), but it offloads the certification burden to every publisher in the catalogue rather than to a single system-wide toggle. SPECULATION: The 16-month delay may reflect Nintendo choosing this whitelist engineering + per-title certification path rather than a system-wide flag — meaning the backlog is not just about one feature arriving late, but about a policy that keeps most of the catalogue VRR-disabled indefinitely.

Adversarial check: the case that this story is wrong

Strongest counter-case: (1) The RTD2175N and RTD2175P are not the same chip, so "the same silicone that gated both consoles" overstates a family resemblance; (2) The Verge's Steam Deck test proved the dock's converter chip could carry VRR, but the console's full video pipeline (APU → dock) is a separate matter, so one cannot definitively prove the launch-day Switch 2 system could have shipped VRR without risk; (3) every outlet already reported the 16-month delay, so this is not novel news.

Response to the counter-case: (1) Acknowledged — the article frames it as the same RTD2175 family, explicitly notes the parts are not interchangeable, and bases its claim on "shared third-party display-bridge silicon," which is accurate; (2) correct, which is why the timeline is labeled a reconstruction and the "capable all along" conclusion is an inference from the Steam Deck test, not asserted as Sony/Nintendo-confirmed fact — and the fact that the feature ships via a dock firmware update (no hardware revision) is itself strong evidence the silicon path was always sufficient; (3) the news is covered, but no major outlet has assembled the cross-console RTD2175-family pattern plus the whitelist-economics divergence — that synthesis is the original contribution.

Adjudication: The story survives. The news event is N1 (well reported), but the cross-console silicon-family pattern and its whitelist-economics reading are N3 (pieces exist in separate forums/threads; the connection appears new).

Why this matters for the console hardware ledger

• A third-party bridge has become a strategic dependency for both Sony and Nintendo. The RTD2175 family's VRR quirks (the DP→HDMI conversion path) are cited in both PS5 and Switch 2 community threads as the same culprit class. That is a supply-chain and product-custody fact about the current console generation that sits outside the custom-GPU narratives.
• The whitelist quietly converts a flagship feature into a per-title maintenance tax. On the PS5, VRR was a system update. On the Switch 2, it is a catalogue-wide patch program that, at day one, zero shipped games had completed. Framed as data: the feature is now "supported" by the platform while the games most likely to need it remain unwhitelisted.
• It reframes the "16 months late" headline. The delay wasn't a hardware miss — it was a chosen rollout architecture. That is the difference between "Nintendo shipped missing hardware" (the common reading) and "Nintendo shipped the hardware, then chose a rollout that defers the benefit to a patch backlog" (the ledger reading).

Confidence and sources

• Primary/near-primary sources: Nintendo Life (Nintendo's own "handheld mode only"/apology statement), Digital Foundry (direct VRR testing of the 23.0.0 dock update, 40–60 Hz / LFC findings), The Verge / Sean Hollister (July 2025 Steam Deck-in-dock VRR proof), WCCFTech (quoting Linneman + the "forces VRR on at all times" quirk), Nintendo patch notes.
• Component-manifest/teardown sources: ResetEra, NeoGAF, gbatemp, kasynparts (RTD2175P/N functions and non-interchangeability).
• Assembly confidence: High. Every factual claim above is observed (component IDs, launch dates, whitelist behavior, Hz range) or labeled as inference. No claim rests on unverified rumor; the "two Switch 2 hardware revisions" claim circulating on YouTube (Nintenduo) was deliberately excluded as unverified.

LostInConsoles investigation. VRR = Variable Refresh Rate. Firmware timing for forward-looking statements is framed as reported; per-game patch adoption is an allocation/outlook, not a confirmed schedule.

Excerpt: In the modern console-APU era — Xenos in 2005, through Durango, RDNA 2, and today — every Xbox GPU was bespoke silicon co-engineered with AMD to hit a power, cost and feature target no PC card matches. A stream of leaked details for Project Helix (2027) now points the other way: the next Xbox's GPU chiplet is off-the-shelf AMD RDNA 5, the "same chip" that will also ship in a desktop Radeon card, with "0 customization on the GPU side," per a leaker Microsoft has otherwise engaged with. No outlet has assembled the structural consequence: it ends a 20-year run of exclusive Xbox GPU architecture, makes Xbox the first console to literally share its GPU die with a desktop card, and splits the console industry's silicon philosophy down the middle — Xbox off-the-shelf and chiplet-scaled, Sony (PS6 Orion) reportedly bespoke and monolithic.

An original LostInConsoles data investigation into the end of bespoke Xbox silicon and the first hardware-generation split in console chip philosophy.

The short version: Every Xbox GPU since the console's modern era has been custom silicon — Xenos (2005) was even designed around an exclusive 10MB of embedded DRAM to eliminate bandwidth as a bottleneck. Project Helix, the 2027 next-gen Xbox, reportedly throws that away: known leaker KeplerL2 said on NeoGAF that Microsoft has "0 customization on the GPU side this time." Corroborating leaks from Moore's Law is Dead describe the Helix GPU chiplet as "literally the same chip that will be used to make the successor to the 9070 XT" — meaning the console's GPU die will be shared with a desktop Radeon card. The CPU and NPU chiplets remain custom; only the GPU goes off-the-shelf. In the same window, leaked PS6 "Orion" specs describe the opposite choice: a monolithic, bespoke APU. Xbox's chiplet-scaled, desktop-die-sharing Helix vs Sony's monolithic, bespoke Orion is the first time in the chiplet era that the two console giants have adopted structurally different silicon custody — and it quietly ends the last thing that made an Xbox an Xbox at the chip level.

Every Xbox GPU up to now was bespoke. That's the point.

The historical baseline matters because it's what "0 customization" breaks. Consoles have almost always used custom GPU designs:

• Xenos (Xbox 360, 2005): a custom ATI design "specifically designed for the Xbox 360" — most famously with 10MB of embedded DRAM (EDRAM) that let it do 4× multisampled 720p with no external memory bandwidth penalty. Nothing in the PC graphics world shipped that part.
• Xbox One / One X (Durango / Scorpio): custom AMD GCN-derived GPUs with console-specific features; the One X's "Scorpio Engine" was a heavily tuned bespoke die.
• Xbox Series X|S (2020): custom RDNA 2 (Oberon/Arden), 12 TFLOPs on Series X, designed with AMD specifically for a fixed ~300W console power envelope and 4K/120 targets. The RDNA 2 architecture powered the RX 6000 desktop cards, but the console variants were custom dies with distinct configs — not the same silicon.

Even Microsoft's original Xbox (2001) used a modified GeForce 3 (NV2A) that NVIDIA "co-developed" with Microsoft, customized for a single-purpose console. So since Xenos, the pattern is clear: every Xbox shipped with console-exclusive GPU silicon. That 20-year run is now reported to end with Project Helix.

The leak: "0 GPU customization," and "the same chip" as a desktop Radeon

The core claim traces to KeplerL2 on NeoGAF (April 2026), a hardware leaker with a "long track record of accurate leaks" (VICE). Paraphrased across WCCFTech, Tech4Gamers/Overclock3D, igor'sLAB, GamingBolt, and ResetEra discussion:

• Microsoft has "0 customization on the GPU side this time" for Project Helix.
• The Helix GPU runs on AMD RDNA 5, using AMD's FSR "Diamond" upscaling/frame-generation models.
• Microsoft's "Magnus" silicon keeps custom CPU and NPU chiplets (reportedly 3× Zen 6 + 8× Zen 6c cores); only the GPU is off-the-shelf.

Then Moore's Law is Dead (MLID) — cited repeatedly for accurate console-silicon scoops — pushed it further: the GPU chiplet inside Helix/Magnus is "literally the same chip that will be used to make the successor to the 9070 XT next year." That is the difference between "using AMD off-the-shelf RDNA" and "using the exact desktop Radeon die." Console and desktop sharing one physical GPU die, at identical config, is effectively unprecedented at this scale.

AMD CEO Lisa Su publicly signaled the semi-custom SoC (codename "Magnus") is "progressing" toward a 2027 launch — shifting the timeline from rumor to plausible. And reports of a ~50-million-unit AMD chip order for the generation, plus a leak describing Magnus as a two-chiplet design totaling ~408mm² on TSMC N3P — the largest APU ever in a console, ~13% larger than Series X — line up with a desktop-scaled, die-sharing design rather than a console-shrunk custom part.

What "family of devices" has to do with it

Project Helix was confirmed (The Verge; Xbox CEO Asha Sharma, Aug 2026) as a "family of devices," not a single box — tiers analogous to Series S/X, including OEM-produced versions and a handheld running a scaled-down Magnus chip that can play both console and PC games. This is where the economics bite: a console-only bespoke GPU die is expensive to amortize at console-alone volume. The LinkedIn/analyst reading of the Magnus leak puts it plainly: by sharing the GPU die with desktop RDNA 5 products, Microsoft avoids the trap that killed its handheld silicon ambitions — insufficient volume to justify semi-custom GPU silicon.

In other words, the "0 customization" GPU is not a downgrade or a corner-cut in raw power; it's a portfolio play. The same AT2-class die can be binned into a desktop mid-range card, a console chiplet, and (scaled) a handheld — one non-recurring-engineering (NRE) expense, three product categories, plus the 50M-console order to drive desktop volumes that AMD itself benefits from. AMD wins (it sells RDNA 5 chiplets into both the console and PC lines), Microsoft wins (no bespoke-GPU NRE), and the loser is the concept of console-exclusive silicon.

Sony is doing the opposite — and that split is the story

The same leak season puts Sony on the other side. Leaked PS6 "Orion" specs (Playground.ru, echoing insider reports) describe a monolithic ~280mm² APU on TSMC 3nm — bespoke, single-die, console-exclusive silicon in the PlayStation tradition. (A competing Insider Gaming report floated chiplet/Navi 5 elements and even a "Canis" PS6 handheld; the monolithic-Orion reading is the better-corroborated one as of this writing, and either way it does not point to a shared desktop-die approach on the Xbox scale.)

So 2027 gives us the first generation where the two hardware giants split on silicon custody:

Whether or not every specific figure holds, the structural divergence is consistent across independent sources and is not assembled anywhere we found.

What this proves — and what it doesn't

Let's be disciplined with the three-layer model.

• OBSERVED FACT (leak-level, not confirmed): KeplerL2's on-record claim (April 2026, NeoGAF) that Project Helix has "0 customization on the GPU side"; MLID's claim that the Helix GPU chiplet is the same die as the 9070-XT-successor class card. Both are secondary/leaker sources, cited and repeated but not confirmed by Microsoft or AMD.
• OBSERVED FACT: Microsoft confirmed Helix is a "family of devices" (Asha Sharma, Aug 2026); AMD's Lisa Su publicly signaled a 2027 semi-custom Xbox SoC ("Magnus").
• INFERENCE (analysis, flagged): The off-the-shelf GPU + family-of-devices configuration implies a die-sharing/NRE-amortization strategy across desktop+console+handheld.
• INFERENCE: PS6 (Orion) monolithic bespoke APU stands in contrast to Xbox's shared-die approach.
• SPECULATION: That this "ends the last thing that made an Xbox an Xbox" as identity loss is editorial framing, not measurable fact — the CPU/NPU remain custom, and Microsoft may still differentiate at the software/API (FSR Diamond, DirectX) layer.

Alternative explanations (we looked)

1. Leaker noise / misquote. The "0 GPU customization" could be imprecise — Microsoft could still tune clock speeds, TDP, or RT configs within a stock RDNA5 die. We treat it as a directional pointer, not a hard spec.
2. It's a "near-custom" part, not stock. The GPU could be desktop-scaled but with console-specific power/clock bins — still a departure from a fully bespoke console die, but not as absolute as "stock card chip." Our framing ("first to share its die with desktop") survives even this softer reading.
3. PS6 also goes chiplet, collapsing the "split." The Insider Gaming chiplet-flavored Orion rumor is real; we've flagged the conflict and leaned on the stronger monolithic reading, but the die-divergence thesis depends on the stronger reading.

Falsification rounds

• Test 1 — "Is there a precedent for a console using a stock GPU die?" The original Xbox's NV2A was a modified GeForce 3, but it was co-developed/customized and never shipped as a desktop card under that config; Xenos, Durango, and RDNA2 were all custom console dies. No modern console shipped the identical die as a desktop card. Standing.
• Test 2 — "Is Helix's GPU actually bespoke after all?" AMD's own language (Analytics Insight) called it a "custom AMD system-on-chip," but that refers to the SoC package — the leaked claim is specifically that the GPU chiplet within is stock. CPU/NPU custom + GPU stock is consistent with both statements. Standing.
• Test 3 — "Did Microsoft debunk KeplerL2?" Microsoft responded to (and "debunked") a third-party/OEM-console rumor and a disc-drive rumor — not the GPU-customization claim. The GPU claim survived the debunk wave. Standing.

Novelty

The raw facts are individually reported (N2/N3): KeplerL2, MLID, WCCFTech, Overclock3D, igor'sLAB, GamingBolt all covered the "0 custom GPU" angle last spring. What we could not find anywhere is the assembled structural story — that Helix marks the first console to literally share its GPU die with a desktop card, and that it creates a generation-defining silicon-philosophy split (Xbox die-sharing/family-of-devices vs Sony bespoke/monolithic) never before seen between the two giants. ResetEra's own thread title ("abandoning the last thing that made an Xbox feel like one") gestures at the emotion but no outlet has done the die-reuse economics + historical-baseline analysis. Novelty: N3.

Confidence

• Core structural claim (Helix GPU is off-the-shelf, ending bespoke Xbox GPU): reported by a reliable leaker + corroborated by MLID + surviving Microsoft's debunk wave — but unconfirmed. Confidence 72.
• PS6 monolithic contrast: 74 (monolithic-Orion is the stronger of conflicting reads).
• Combined "industry split" thesis: 70 — publishable as a clearly-flagged investigation of leaked/under-assembled architecture signals, not as confirmed fact.

LossInConsoles verification: All figures above are attributed to, and quarantined as, leaker/reporting sources; nothing is presented as confirmed by Microsoft, AMD, or Sony. The die-sharing and architecture-divergence are flagged as INFERENCE/ANALYSIS. This is the first independent treatment connecting the "0 GPU customization" leak to the end of bespoke Xbox silicon and the Xbox-vs-Sony custody split.

Excerpt: In March 2026 NVIDIA quietly restarted RTX 3060 production on Samsung's 8nm process — the exact node that fabricates the Nintendo Switch 2's T239 SoC. Five months later, Samsung's 8nm wafers repriced roughly 10% for new orders, effective September 2026 — the same month Nintendo imposed the first post-launch price increase in its history, to $499.99. Neither the PC press nor the console press has connected these two facts: NVIDIA is now simultaneously the customer occupying Nintendo's "safe" mature node with a revived GPU, and the supplier is re-pricing that node the month Nintendo broke its price discipline.

An original LostInConsoles data investigation into the quiet collision on Samsung's 8nm foundry line — two NVIDIA products, one node, one month.

The short version: The Switch 2's T239 SoC lives on Samsung 8nm (SF8) — the node our own earlier reporting called the console's insulation from the AI-capacity crunch. That insulation is eroding from two directions at once. First, NVIDIA resurrected the RTX 3060 (also Samsung 8nm, also Ampere) in March 2026, becoming a second, concurrent occupant of the same mature line that supplies the console — both products are NVIDIA's. Second, Samsung's foundry repriced 8nm up ~10% for new orders, effective September 2026 per reporting, in the same month Nintendo's first-ever post-launch price hike ($449.99 → $499.99) went live. The node that was supposed to be "the one foundry not being rationed to AI" is now actively re-pricing Nintendo's core silicon, at the worst possible moment.

The node the whole industry said was safe

When the Switch 2 launched June 5, 2025, its custom NVIDIA Tegra T239 system-on-chip was fabbed on Samsung's 8nm mature process (SF8). That attribution is confirmed by die-shot analysis (TechPowerUp), NVIDIA/Nintendo shipping manifests, Linux disclosures, and Digital Foundry's hardware teardown — it is not in dispute.

The console press read that as a cost-cutting compromise: an older, less power-efficient node than the 4nm/5nm parts in flagships. Our analysis last month offered a different reading — that Samsung 8nm was the one mature foundry not being rationed to AI, which made it a structural insulation play in a semiconductor market where advanced-node wafers go to the highest-margin AI customer while TSMC raises prices across 3nm/2nm annually.

That thesis now needs a serious markdown. The "safe" node is no longer idle, and someone has put a meter on it.

NVIDIA came back to the same line — with a GPU that competes for the same wafers

In March 2026, as AI-buildout-induced GPU shortages squeezed the consumer market, a cluster of reporting (Videocardz citing South Korea's Hankyung, igorslab, PCMag, WCCFtech, TechPowerUp, PauseHardware) confirmed NVIDIA would revive the GeForce RTX 3060 — based on the Ampere GA106 chip, which was originally manufactured on Samsung's 8nm process. Samsung's foundry division reportedly restarted its 8nm DUV lines to restart the card, adding roughly 30,000–40,000 wafer starts per month (about 10% of Samsung's total capacity, per sector-derived figures widely echoed).

The RTX 3060 and the Switch 2's T239 are both Ampere, both on Samsung 8nm, both designed and bought by NVIDIA. NVIDIA is on both sides of the same node: it is the console's SoC customer, and it is the revived-GPU customer reclaiming the same mature-line capacity to sell a $330 card.

That coupling — one supplier, one node, one customer, two competing product lines, one of them a 24-million-unit-a-year console — is the connection no outlet has assembled. The console that was supposed to be insulated is now sharing its foundry home with a mass-market GPU whose whole reason to exist is that the same AI-driven shortage is pushing consumers to old cheap hardware.

And then Samsung repriced the node the month Nintendo raised its price

The second, harder signal is the price meter. In late August 2026, Reuters reported — citing two people familiar with the matter — that Samsung raised prices on advanced contract chipmaking by up to 15% for new orders as AI demand tightened capacity. The breakdown, echoed by Global Sources, Tom's Hardware, Yahoo Finance and Trend Hunter: 4nm and 5nm rose about 10–15%; the older 8nm node rose about 10%. Global Sources explicitly dated the increases as taking effect "next month, in September 2026."

Now lay that against the console calendar. Nintendo's first-ever post-launch price increase — from $449.99 to $499.99 — took effect September 1, 2026. (Prior to this, Nintendo had never raised a console's price after its launch; our September 3 investigation tracked the parallel display and battery re-engineering that accompanied it.)

The sequence is: Nintendo raises the console $50 on Sept 1, then Samsung's 8nm repricing — the node under the T239 — takes effect that same September. Several outlets (TwistedVoxel, Tech4Gamers, OtakuKart, TweakTown, inf.news) have each individually flagged that the 8nm increase "may be on the cards" for another Switch 2 price action. Nobody has put the node-occupancy and node-repricing facts in the same story — and that combination is what changes the read.

What this actually proves — and what it doesn't

Let us be careful with the three-layer distinction.

Observed (high confidence):

• Switch 2 SoC (T239) is Samsung 8nm. Confirmed.
• NVIDIA revived the RTX 3060 on Samsung 8nm. Confirmed by multiple outlets citing industry sources.
• Samsung repriced 8nm ~10% for new orders, effective ~September 2026 (Reuters + Global Sources dating). Confirmed.
• Nintendo's price hike to $499.99 took effect September 1, 2026. Confirmed.

Inference (reasonable from the above):

• NVIDIA is a concurrent, two-sided occupant of the 8nm node that supplies the console, and Samsung 8nm — the "safe" node — is now actively re-priced in the same month Nintendo raised the console's price. Reasonable.
• The 8nm repricing lands on the T239's foundry cost at the worst possible time for Nintendo's margin. Reasonable.

Speculation (do not present as fact):

• That a second Switch 2 price hike is imminent. Nothing is announced; this is trade-press reading, and we flag it as such.
• That NVIDIA (the foundry contract holder for T239) rather than Nintendo absorbs the 8nm increase. The buyer of record is NVIDIA, so contract structure is unknown.
• That the RTX 3060 outright steals Switch 2 wafer allocation. Samsung's 8nm lines reportedly had idle slack — which is precisely why the 3060 revival was feasible. The volume-competition story is weaker than the price story.

The durable, defensible claim is not "a second Switch 2 price hike is coming." It is: the structural protection the console's mature node was said to provide is eroding, observably, in the exact month Nintendo broke its pricing history — and the agent of the erosion is NVIDIA itself, which sits on both sides of the same line.

Why the gaming press missed it

The PC-hardware beat covered the RTX 3060 revival as a GPU-shortage story. The console beat covered the Switch 2 price increase as a Nintendo-cost story. Reuters covered Samsung's repricing as a foundry-economics story. Each is correct and each is incomplete, because the overlap is not in any single feed: one foundry node, one customer (NVIDIA), two of that customer's high-volume products, repriced the month a console raised its price. Assembling that ledger is a data-connection, not a news-summary — and it is the connection that has not been made anywhere in the sources we searched.

The broader point for the console industry: the era when a mature node was a low-cost, low-drama refuge is closing. Samsung's advanced 4nm capacity is more than 50% dedicated to HBM4 base dies and reportedly booked through 2027; NVIDIA, OpenAI and Tesla are all pulling on Samsung capacity; and the company raised prices while losing foundry share — a sign its lines are genuinely full on the advanced side, and it is re-pricing even the mature node the Switch 2 depends on. The console's protection was never a property of the node — it was a property of nobody else wanting it. That is no longer true.

Confidence levels: Foundry node & repricing facts: HIGH (Reuters + multiple corroborations). RTX 3060 revival: HIGH. Nintendo price-hike timing: HIGH. Second-price-hike implication: LOW-MEDIUM (analysis/rumor, clearly flagged). NVIDIA-absorbs-cost: SPECULATION. Capacity-collision magnitude: MEDIUM (idle slack weakens volume element; price element carries).

Sources: Reuters (Aug 19, 2026); Videocardz/Hankyung (Mar 9, 2026); igorslab; PCMag; WCCFtech; TechPowerUp; Digital Foundry (2025); Global Sources; Tom's Hardware; Yahoo Finance; TrendForce; ixbt.com (Sept 7, 2026); TwistedVoxel; Tech4Gamers; OtakuKart; TweakTown; inf.news; tech-insider.org.

An original LostInConsoles investigation. Forward-looking supply and pricing claims are reported/allocation-based; unconfirmed leaks are flagged as rumors; margin and node-coupling synthesis is our own analysis.

Excerpt: Trade policy moved ~75% of US-bound console assembly to a narrow strip of northern Vietnam. The AI-hardware boom moved there too — OpenAI's first consumer devices (Sweetpea, Gumdrop) are assigned to the same Foxconn units in the same provinces that now solder your Xbox and PlayStation. The result is a collision the games press has not connected: consoles and AI wearables are competing for the same factory floor space, the same ~330,000-worker labor pool, and the same ODM priority inside Foxconn. This is the assembly-layer squeeze that sits below the silicon and memory squeezes everyone already reported.

An original LostInConsoles data investigation into the geographic collision between console re-shoring and AI-device manufacturing in northern Vietnam. OpenAI device details are unconfirmed rumors (flagged); Foxconn segment revenue, Vietnam console share, the Bac Ninh labor crunch, and the Fushan/Xbox permits are sourced facts; the collision interpretation is labeled analysis.

The short version

In 2025, tariffs emptied Chinese console production and pushed it to northern Vietnam. Today Vietnam assembles roughly 75% of all US-bound game consoles — Xbox at Foxconn's Fushan unit in Bac Ninh, PlayStation at Foxconn in Quang Ninh, Switch 2 alongside them in the north — while China's share of US console imports fell from ~86% to under 5% (Bloomberg graphics, Dec 2025).

In that same calendar window, the AI-hardware boom reached the exact same patch of map. OpenAI's first consumer devices — the "Sweetpea" audio wearable and the "Gumdrop" pen-style device, with a portfolio of up to five devices assigned to Foxconn by Q4 2028 — are reported to be built by Foxconn in Vietnam, the same corporate family and the same provinces that now assemble the consoles (duckittech, biggo, hackwarenews; unconfirmed).

Two massive re-shoring waves did not choose different countries. They chose the same assembly corridor — and nobody in the games press has connected the two.

The evidence: two re-shorings, one corridor

Console re-shoring (OBSERVED)

• Vietnam ~75% of US-bound consoles, 80%+ China share collapse. Bloomberg (Dec 10, 2025) documented China's share of US video-game console imports falling from 86% to under 5%, with Vietnam taking the rest. Corroborated by thailand-business-news, digitalinasia.
• Xbox → Fushan Technology (Foxconn), Bac Ninh. Reuters (Dec 2025) reported Fushan Technology (Vietnam) LLC submitted an environmental permit to expand Bac Ninh production to include Xbox consoles and UAVs (~$316M, VND8,354bn project). Windows Central puts the targeted scale at up to 4.8M additional Xbox devices/yr; TWSE filings show an additional $58.32M into Fushan (Apr 2026). Foxconn EV Energy & Component has capacity to produce up to 4M gaming devices in northern Vietnam (company doc, 2024).
• Luxshare → Xbox/gaming consoles in Nghe An and Bac Ninh (Reuters; ~4.5M units/yr target per permits).
• PlayStation → Foxconn, Quang Ninh. Foxconn received an investment registration for ~$287.2M PlayStation console production; Foxconn Singapore holds a $263.7M "smart entertainment products" licence with ~4.18M units/yr capacity; plus $551M more approved in Quang Ninh (theinvestor.vn, Reuters).
• Switch 2: north Vietnam production reported across multiple outlets.

AI-device re-shoring (REPORTED / unconfirmed)

• OpenAI Sweetpea (audio wearable) & Gumdrop (pen device) — screenless, voice/handwriting-to-ChatGPT, iPhone-class BOM, $250–400. Reported via Taiwan/industry pipeline, not confirmed by OpenAI, Foxconn, or Samsung.
• Manufacturing assigned to Foxconn in Vietnam, after OpenAI reportedly moved the project from China-based Luxshare to Foxconn (quasa, ai magazine, techplustrends).
• Production scale: Sweetpea est. 40–50M units in first year; Foxconn told to prepare for 5 total devices by Q4 2028 (hackwarenews, introl). Conflicting "late 2026" vs "Q4 2028" launch windows.

The collision (INFERENCE — the original synthesis)

Consoles and AI wearables don't merely share a country. They share:

1. The same OEM family — Foxconn assembles Xbox (Fushan/Bac Ninh), PlayStation (Quang Ninh), AND the OpenAI devices.
2. The same provinces — Bac Ninh, Bac Giang, Quang Ninh.
3. The same ODM priority queue inside Foxconn, which has publicly flipped to AI-first, and
4. The same labor pool, which is already running out.

The labor constraint that makes the collision real

Bac Ninh — the province where Xbox assembly lives — cannot staff its electronics parks:

• 330,000+ factory workers needed in Bac Ninh in 2026, with companies in a "signing bonus race" so intense that workers began gaming the system to cash multiple bonuses (VnExpress, Mar 14, 2026).
• Broader northern electronics cluster (Bac Ninh/Bac Giang) reported needing up to 334,000 workers and recruiting foreign workers for the first time as domestic supply runs dry (briefasia, vnrobo, Mar–Jul 2026).
• A December 2025 job fair filled fewer than 10% of 35,000 open positions (briefasia, Jul 2026).
• Recruitment bonuses reaching up to ~two months' salary in Bac Giang/Bac Ninh/Nghe An factories (vnrobo).

Why this matters for consoles: console assembly is among the most labor-intensive steps in electronics — final assembly, testing, packing are not easily automated at tens-of-millions scale. An ODM running three shifts is more constrained by people than by wafers. When OpenAI wants 40–50M devices out of the same corridor in year one, it is not competing for TSMC capacity (that was the 09-09 story) — it is competing for assembly bodies in Bac Ninh.

INVESTIGATION VALUE framing (analysis): the silicon squeeze (09-09) and memory squeeze (prior stories) both constrain consoles. But this investigation is about the third, unmapped layer: geographic/assembly-floor collision. The console's problems are no longer just upstream (chips, memory) — they are becoming co-located with a competing high-growth product line.

Foxconn's priority flip is the visible mechanism

• Q2 2026: Hon Hai net profit NT$59.98B (~$1.86B), +35% YoY, a record for the quarter (emsnow, AppleMagazine).
• Q2 2026 revenue ~NT$2.5 trillion, +40% YoY (theoutpost, powerdrill).
• Cloud & networking (AI servers, racks, switches) became the top segment — reportedly past 50% of revenue, while the smart consumer electronics segment (where consoles sit) fell to ~29% (AppleMagazine, powerdrill breakdowns).
• The console-making division is now a minority, lower-margin segment inside the world's largest contract electronics maker.

Alt explanation (considered): Apple's iPhone business also lives in the same Foxconn consumer-electronics segment, so the 29% is not console-only. True. But the point stands directionally: Foxconn's capital, capacity, and hiring priority have rotated toward AI — and consoles ride in the segment that lost its top billing. (Analysis.)

Alt explanation (considered): Console output may not actually be shrinking — Microsoft/Foxconn are expanding Xbox lines in Vietnam (+4.8M target). Expansion and squeeze are not mutually exclusive: the ODM is adding lines while also telling the same region it needs hundreds of thousands more workers. The binding constraint is the shared labor pool and floor space, not the line count.

The part nobody has connected

Gaming press covered "Xbox is now made in Vietnam" (Dec 2025) and "OpenAI's first device goes to Foxconn" (Jan 2026) as separate supply-chain stories. No prior reporting was found connecting the two into the collision claim: the same ODM corridor that was chosen to rescue console supply from tariffs is simultaneously the chosen home of the AI wearables that are about to out-bid it for labor, floor space, and ODM priority. That connection is the original contribution of this report.

Two secondary threads worth logging:

1. Foxconn is pulling Apple audio assembly in-house in Vietnam (Counterpoint-reported Foxconn investment in Goertek's new Vietnam unit) — more hiring demand in the same corridor (inf.news).
2. OpenAI's devices sit above consoles in Foxconn's queue by economics — 40–50M units of a $250–400 device out-earns a $500 (or, post-2026, ~$650–800) console on volume-weighted gross margin. (Analysis.)

Verdict

Confidence: 72/100 — STRONG enough to publish, with clear flags.

What is OBSERVED (cited): Vietnam ≈75% of US-bound consoles & China share collapse (Bloomberg); Xbox→Fushan/Foxconn Bac Ninh (Reuters/TWSE/Windows Central); Playstation→Foxconn Quang Ninh (theinvestor/Reuters); Bac Ninh labor shortage 330k+ & signing-bonus race (VnExpress); Foxconn Q2 2026 +35%/-40% record results with AI-first mix (emsnow/AppleMagazine/powerdrill).

What is REPORTED/UNCONFIRMED (flagged): OpenAI Sweetpea/Gumdrop device specs, Foxconn-in-Vietnam assignment, 40–50M-unit target, 5-device-by-Q4-2028 plan (hackwarenews, duckittech, introl, biggo); Luxshare being dropped for Foxconn (quasa).

What is INFERENCE (analysis): that console and AI assembly collide in the same corridor; that the shared labor pool and ODM priority are binding constraints; that OpenAI devices outrank consoles in Foxconn's internal queue.

What this does NOT prove: It does NOT prove consoles are already shortage-constrained by labor (retailer GTA 6 warnings cite DRAM/component prices, not labor). It does NOT prove Foxconn is reducing console output. It proves the collision exists and the pressure vectors — labor, floor space, ODM priority — all point the same direction and no one has mapped it.

Falsification: We searched specifically for any source connecting console re-shoring to OpenAI-device co-location/labor competition in northern Vietnam. None found. Searches for "console production delayed by Vietnam labor shortage" and "OpenAI device competes with Xbox for assembly" return no direct reporting — supporting novelty.

Novelty: N3 — the pieces exist (Vietnam console share; OpenAI-Foxconn-Vietnam; Bac Ninh labor crunch) but the connection is new.

Alternative explanations (adjudicated): (a) console output actually expanding (addressed — expansion ≠ no squeeze on shared floor/labor); (b) OpenAI device is vaporware and never scales (risk — but even a fraction of 40–50M units is material to the corridor; flagged); (c) labor is only a cost issue, not a capacity issue (refuted by 330k-worker gap and unstaffed job fairs).

Fact discipline

OBSERVED FACTS (cited): China console-import share 86%→<5% & Vietnam ~75% of US-bound (Bloomberg, 2025-12-10); Fushan/Xbox Bac Ninh expansion permit ~VND8,354bn & +$58.3M (theinvestor, Reuters, inf.news, 2025-12/2026-04); Xbox +4.8M/yr target (Windows Central); Foxconn Quang Ninh PlayStation $287.2M & 4.18M smart-entertainment units/yr & $551M (theinvestor, Reuters, businesstimes); Luxshare gaming consoles ~4.5M/yr (Reuters); Bac Ninh >330,000 workers needed, signing-bonus race (VnExpress, 2026-03-14); northern electronics 334,000 workers, foreign-worker recruitment, <10% of 35k job-fair positions filled (briefasia, 2026-07; vnrobo); Hon Hai Q2-26 net profit NT$59.98B +35%, revenue ~NT$2.5T (emsnow, theoutpost, powerdrill); Foxconn cloud/networking >50% revenue & consumer-electronics ~29% (AppleMagazine, powerdrill, 2026).

REPORTED (flagged as rumors): OpenAI "Sweetpea" & "Gumdrop" devices, Foxconn-in-Vietnam assignment, 40–50M-unit first-year target, 5 devices by Q4 2028 (hackwarenews, duckittech, introl, biggo, quasa — all unconfirmed by OpenAI/Foxconn); Luxshare dropped for Foxconn (quasa).

INFERENCE (analysis, labeled): the same-corridor collision; shared labor/floor/priority as binding constraints; OpenAI devices outranking consoles in Foxconn's economic queue.

This is an original LostInConsoles investigation. It does not constitute investment advice. No claim is made that OpenAI, Foxconn, Microsoft, or Sony have confirmed any leaked device detail.

Excerpt: Every next-generation console in the modern era — PS5, Xbox Series X, PS4, Xbox One, PS3, Xbox 360 — launched on the most advanced chip-manufacturing node available at the time. PS6 ("Orion") and Xbox ("Magnus"/Project Helix) break that streak. Both are reported to be built on TSMC's 3nm N3P process, not the 2nm leading edge, even though they launch in 2027 — more than a year after 2nm entered volume production. Behind the node choice is a structural change the industry has never priced in before: AI and datacenter buyers (Apple, Nvidia, AMD, OpenAI, Microsoft) have booked out TSMC's 2nm capacity so completely that a device shipping at ~$500–$1,000 per unit — and selling tens of millions — no longer ranks for leading-edge wafers. The console is becoming a second-class citizen of the semiconductor industry at exactly the moment its components got most expensive.

An original LostInConsoles data investigation into why the 2027 generation is the first ever to skip the leading-edge node — and why 3nm is now the most expensive node the console business has ever depended on. All forward-looking and unconfirmed node/price details are explicitly flagged; capacity and pricing facts are cited. The structural interpretation is labeled as analysis.

The short version

For six console generations, the rhythm was the same: wait for the newest process node, then launch the flagship console on it. It was a near-absolute rule of the hardware business — gaming was big enough, and early enough in a node's life, that console silicon got leading-edge wafers. The 2027 generation breaks that rule.

Here is the reported state of play for the next two machines:

• PS6 ("Orion"): TSMC 3nm, AMD Zen 6 CPU cores, RDNA 5 GPU (52–54 CU, ~2.6–3.0/3.6 GHz), ~40 GB GDDR7 (30/32 GB in some threads), monolithic 3nm die. Reported via multiple insider pipelines (MLID-sourced) — unconfirmed by Sony.
• Xbox ("Magnus" / Project Helix): dual-chiplet — a 144 mm² SoC die on TSMC N3P (CPU, NPU, I/O) plus a 264 mm² GPU die on N3C/N3P. Reported via the same leak pipeline — unconfirmed by Microsoft.

Neither design is on 2nm. Both are on the N3 family. And while it is tempting to write that off as a cost decision — 3nm is cheaper per wafer than 2nm — the deeper story is that consoles couldn't get 2nm even if they wanted it. TSMC's advanced-node capacity is booked out through 2028. Apple alone reserved more than half of the initial N2 allocation. Nvidia, AMD, Qualcomm, Google, OpenAI, and Microsoft's own AI division have claimed the rest. Consoles, which launch in volumes of tens of millions and need cost predictability, do not belong to that club anymore.

This is the first time in the console generation cycle that "leading-edge silicon" and "console silicon" are two different things — and one of them no longer gets the good wafers.

The evidence: next-gen is on 3nm

The two next-gen leak pipelines have been unusually consistent on the node.

PS6 "Orion" (3nm). The most detailed public spec dump — from Moore's Law is Dead, relayed by multiple outlets — describes a monolithic 3nm die: 10 Zen 6 cores, an RDNA 5 GPU at 52–54 compute units, up to ~40 TFLOPS, and ~40 GB of GDDR7. Every rendition, from Russian-language roundups (playground.ru, podpiska.org, dzen) to Western coverage (geeky-gadgets, wccftech), puts the SoC on TSMC 3nm. None put it on 2nm.

Xbox "Magnus" (N3P). A separate insider stream describes a dual-chiplet part: a 144 mm² SoC die fabricated on TSMC N3P holding the CPU cores, NPU, and main I/O, plus a 264 mm² GPU die on N3C or N3P. Notebookcheck, fandomwire, alltechnerd, gsmgotech, and a ResetEra leak-compilation thread all converge on the same node.

Two independent insider pipelines arriving at the same process family is meaningful. Neither manufacturer has confirmed it, and both have every reason to stay quiet, so these are reported facts, not confirmed ones — but they are the only credible signal we have, and they point in one direction: 3nm.

Falsification note: we searched specifically for any report that Sony or Microsoft had moved PS6/Magnus to 2nm. There is none. If such a switch were in play, the leaks that are so detailed elsewhere would almost certainly reflect it. They do not.

Why 2nm isn't available: the booking ledger

The reason consoles land on 3nm is not a lack of ambition — it is a lack of wafers. TSMC's leading edge has been sold to AI and computing customers years in advance:

• "TSMC's advanced chip capacity is fully booked through 2028." Nvidia, Apple, AMD, and Qualcomm have reserved essentially all 2nm slots; 3nm has been at full utilization since it entered mass production (dataconomy.com, March 2026).
• Apple took more than half of the initial N2 allocation for the A20 (iPhone 17 family) and M6 (Mac) silicon — the largest single-customer allocation on a TSMC leading-edge node since the original N3 ramp (presenc.ai).
• 2nm is scaling fast. N2 entered volume production in Q4 2025 and now has "four times as many tape-outs as 3nm" at the same stage (wccftech). TSMC aims for ~100,000 2nm wafer-starts per month by end-2026, with combined 2nm + 3nm exceeding 400,000 wafers by 2028 (TrendForce, Sept 7 2026).
• Yield problems made it worse. 2nm volume production slipped to late 2026 on yield issues, which kept the advanced nodes scarcer still — and pushed Apple/Nvidia to grab even more upfront.

When every 2nm wafer is already sold to buyers who pay datacenter-grade prices and book capacity in tranches of tens of thousands of wafers, a console launch — which must hit a mass-market price point years in advance — cannot compete for that allocation. The console becomes the buyer that gets whatever is left.

The catch: 3nm isn't a refuge, it's the new pressure point

Here is the uncomfortable part of the story. If consoles are pushed down to 3nm, they are not landing somewhere cheap and quiet — they are landing on the node that is now the most crowded and most expensive in TSMC's lineup.

TrendForce reported on September 7, 2026 that TSMC's 3nm family is on track to overtake 5nm as the company's top revenue node in the second half of 2026, with 3Q26 output value above NT$400 billion (~30%+ of total revenue). It is capacity-constrained, TSMC is adding a new 3nm fab at Tainan (1H27 mass production), and — critically — TSMC and its customers agreed to a 3nm price hike given the demand (wccftech). At roughly $18,000 per wafer, N3P is among the most expensive nodes ever put into a mass-market console.

What is crowding that node? Not just the consoles — every player that couldn't get 2nm either. OpenAI's first-generation Jalapeño chip (co-developed with Broadcom) is built on 3nm. Nvidia's Rubin family and Google TPUs are on or moving toward the N3 family. So the AI industry that pushed consoles off 2nm is now also sitting on 3nm — meaning the "fallback" node for next-gen consoles is exactly as contested as the leading edge, just at a slightly older process.

The result: the 2027 generation is spending the most it ever has on silicon while getting the second-best node — a double squeeze no console generation has faced before.

The precedent that broke: every prior generation launched on the leading edge

This is the part that makes 2027 genuinely different, and it deserves to be stated plainly.

• PS5 / Xbox Series X (2020): TSMC 7nm — the leading node of their time.
• PS4 Pro / Xbox One X mid-gen flagships and the base PS4 / Xbox One (2013): 28nm — leading edge at launch.
• PS3 / Xbox 360 (2005–2006): leading-edge 65nm (after 90nm launch), the newest available Silicon-On-Insulator process.
• PS2, PS1, N64, SNES, NES: each deployed the most advanced commercial silicon available at their console debut.

For decades, a flagship console was a reason for a foundry to ramp a new node — Sony and Microsoft used to be anchor customers. Today, at 3nm, that role has been taken over by Apple, Nvidia, OpenAI, and Google. The console no longer defines what the new node means; it gets admitted to the previous one, and pays a premium to be there.

Inference (flagged as analysis): this is a structural demotion of the console from the leading edge, driven by AI's claim on advanced capacity — not a choice Sony and Microsoft are making from a position of strength.

What it means for price and for the market

The hardware realities collide with a market that is already warning about sticker shock:

• AMD's gaming segment fell 31% year-over-year to $779M in Q2 2026, driven by lower semi-custom console sales and weaker discrete GPU demand — the console cycle is at a trough right before the next-gen ramp.
• Next-gen consoles at ~$1,000 would cut first-five-year sales by roughly 38%, Ampere Analysis warned (via wccftech and other outlets). Reported PS6 hardware cost estimates put the BOM near $960 (noobfeed) — a figure that would all but force a near-$1,000 price if subsidized on silicon costs alone.
• GDDR pricing spiked again in the second half of 2026 (memory-crisis headlines have tracked +10–30% price moves at retail).

Reported vs. confirmed: the $960 BOM, the $1,000 price, the 38% sales-cut estimate, and the exact node assignments are all reported/analyst figures, not confirmed by Sony, Microsoft, or AMD. What is firm is the capacity direction: 2nm is not available to consoles; 3nm is crowded and getting more expensive.

The part nobody has connected

The mainstream press has covered "PS6 is on 3nm" and "2nm is sold out," but almost no one has connected the two into the structural claim: this is the first console generation ever to be demoted from the leading edge, and it is demoted to the exact node the AI giants are now crowding. That is the novel synthesis of this investigation.

Two secondary threads are worth logging even though they are not the headline:

1. NVIDIA revived the RTX 3060 on Samsung 8nm in 2026 — filling Samsung's idle mature-node capacity — which is the same node family as the Switch 2's T239. If AI demand ever pulls up Samsung's 8nm utilization, Nintendo's only non-TSMC console loses its one insulation from the AI squeeze. (Secondary, speculative — flagged.)
2. Microsoft's own Maia 300 AI accelerator (~300,000 units requested from TSMC for 2027) sits in line ahead of Xbox Magnus for internal TSMC capacity. Microsoft is effectively competing with itself for wafers next generation.

Verdict

Confidence: 75/100 — STRONG. The structural capacity facts are well-sourced (TrendForce, TSMC, dataconomy, presenc.ai, industry reporting). The specific node assignments are leak-based and flagged as reported, not confirmed. The synthesis is novel and falsifiable.

What this does NOT prove: it does not confirm the final node of PS6/Magnus (Sony/MS could still surprise), does not confirm the $1,000 price, and does not prove that AI capacity alone forced the 3nm choice — cost and die economics are legitimate co-factors. What it does establish is the direction: consoles no longer lead the node race, and the fallback node they occupy is exactly where AI demand has migrated.

Alternative explanations considered: (a) 3nm chosen for cost/die-size/clock targets rather than capacity — plausible co-factor, but it does not explain why 2nm is unavailable; (b) leak inaccuracy — possible, but two independent pipelines agree; (c) consoles deliberately skipping 2nm to save cost — unlikely to apply to Microsoft, whose whole Helix pitch is performance leadership.

Novelty: the individual facts (3nm PS6, 2nm booking, 3nm revenue crown) are all reported. The connection — the first console generation demoted from the leading edge because AI claimed 2nm — was not found in prior coverage and is the original contribution of this report.

Fact discipline

OBSERVED FACTS (cited): TSMC 2nm booking/sold-out through 2028 (dataconomy.com, 2026-03-31); Apple >50% N2 allocation (presenc.ai); 2nm tape-outs 4× vs 3nm and N2 volume production Q4 2025 (wccftech); 3nm to overtake 5nm as top revenue node 2H26, ~NT$400B in 3Q26, capacity-constrained, Tainan fab adding capacity (TrendForce, 2026-09-07); TSMC 3nm price hike agreed (wccftech); 3nm wafer ~$18,000 (Extremetech); OpenAI Jalapeño on 3nm, Nvidia Rubin, Google TPUs on N3 family (TrendForce/press); AMD gaming revenue -31% to $779M Q2 2026 (SEC/AMD); Switch 2 = Samsung 8nm, only non-TSMC current-gen console (08-29 LostInConsoles investigation, teardown-backed); PS5/Xbox = 7nm at launch, PS4 = 28nm, prior-gen history.

REPORTED (flagged): PS6 "Orion" 3nm + specs + 40GB GDDR7 (MLID leak via press — unconfirmed by Sony); Xbox "Magnus" N3P 144mm²/264mm² chiplets (leak via press — unconfirmed by Microsoft); PS6 BOM ~$960 (noobfeed); Samsung 2nm Taylor full booking + customers Tesla/Broadcom/Arm + ~40% 2nm yields (wccftech/industry); Samsung 4nm HBM4 base-die diversion + sold out through 2027 (industry reporting).

INFERENCE (analysis): that the console generation has been structurally demoted from the leading edge by AI's claim on 2nm capacity; that 3nm crowding raises next-gen console BOM; that RTX 3060 revival on Samsung 8nm shares the Switch 2's node. All labeled as analysis throughout.

This is an original LostInConsoles investigation. It does not constitute investment advice and makes no claim that Sony, Microsoft, or AMD have confirmed any leaked specification.

Excerpt: Every retro-handheld maker launched a dual-screen clamshell into 2026 — Anbernic's RG DS and RG DS Plus, AYN's Thor, AYANEO's Pocket DS — the most component-hungry form factor the industry makes, with two panels, more silicon, and more memory per unit than any single-screen handheld. They did it in the worst memory-cost market in a decade, the same environment that pushed the Steam Deck OLED up ~46%. The clearest proof of the collision is hiding inside the product Anbernic calls its "Plus." The RG DS Plus ships with 1 GB of RAM — down from 3 GB on the original RG DS — with the internal eMMC storage removed entirely, officially "to save cost." It is the one part of the industry where a mid-cycle refresh is going backwards on memory on purpose. Behind it is a curious piece of co-development: the Nintendo DS emulator ecosystem that makes these devices worth buying was itself built with test hardware provided by the handheld makers.

An original LostInConsoles data investigation into why the hardware industry's most memory-intensive form factor — the dual-screen handheld — arrived at the worst moment in a decade, and how budget makers are quietly cutting RAM to stay under $100 while the emulator ecosystem co-developed with them. All forward-looking and unconfirmed details are explicitly flagged; price and spec facts are cited. Margin and supply interpretations are labeled as analysis.

The short version

In 2026 the retro-handheld industry collectively revived the last form factor the console world had abandoned. After Nintendo discontinued the 3DS in 2020 and never shipped a successor, dual-screen remained dead for half a decade. Then, within roughly twelve months, nearly every major Android retro-handheld maker shipped a clamshell with two screens:

• Anbernic RG DS (existing) and RG DS Plus (unveiled Sept 4, 2026) — the budget tier.
• AYN Thor — a dual-screen flagship (6-inch 1080p/120Hz OLED top, 3.92-inch 60Hz OLED bottom), shipped in batches throughout 2025-2026.
• AYANEO Pocket DS — the $399 Android dual-screen option (Snapdragon, Indiegogo).
• AYANEO Flip 1S DS (AMD Ryzen AI HX370) at the premium Windows end.

The enthusiast press has covered this as "2026: the year of the dual-screen," and that framing is fair. What hasn't been connected is the economics. The dual-screen handheld is, structurally, the worst form factor a maker could have chosen to launch into the 2026 memory crisis. A dual-screen device does not just carry one more display — it carries a second panel, a second backlight driver, more board routing, typically more RAM (to run both screens and modern Android), and more storage to hold both a big OS and a growing game library. Every one of those is a line item that spiked over the past four quarters. The memory shortage that started with AI data-center demand has pushed LPDDR5 prices up 80-90% in a single quarter (Q1 2026), and NAND storage climbed alongside DRAM. Samsung has warned the shortage could persist until 2028.

The result is a market that split in two on the same form factor — and the budget tier, specifically, is being forced to de-spec a product that is nominally an upgrade.

The dataset: a "Plus" model with one-third the RAM

On September 4, 2026, Anbernic unveiled the RG DS Plus. The official announcement (anbernic.com) emphasized the things that read as a clean upgrade:

• Dual screens grown from 4.0-inch 640×480 on the RG DS to dual 4.3-inch 1024×768 (about 15.6% more screen area), enabling 4× pixel scaling for Nintendo DS.
• A redesigned 15°-180° hinge, a capacitive stylus stored inside the shell, microphone support, and a Linux-first interface ("Screen Swap," lower-screen touch control, real-time clock, instant save states) built around a new in-house DS emulator, NNDSS.

What the official announcement did not list was the memory configuration. As several outlets noted, the reveal was heavy on screens and light on a full spec sheet. When the spec information finally landed, it told a striking story that ran against the "Plus" label:

• RAM: 1 GB — down from 3 GB on the original RG DS.
• Internal eMMC storage: removed entirely (the RG DS had 32 GB of eMMC). System and games now run from the microSD card only.
• The chipset is reported as the Rockchip RK3568 (the same SoC family, not a faster part).

Retro Handhelds' spec breakdown is blunt about the reasoning, in the source's own phrasing: "No eMMC Storage (save cost) — 1GB Ram (down from 3gb — 1gb is all that's needed for GammaOS)." Yanko Design independently reported the same: Anbernic "ditched the built-in eMMC memory to keep the cost the same as the earlier version," and the RAM downgrade to 1 GB is tied to a collaboration with community firmware developer @TheGammaSqueeze to port GammaOS to the device. The justification is that GammaOS runs in 1 GB, so a 1 GB part keeps the price "under $100."

Let me be precise about what this is and is not. This is not an accidental defect — Anbernic's 2026 track record already includes units that shipped with unannounced RAM cuts (the RG34XX SP was quietly cut from 2 GB to 1 GB in January, and some units turned up with 512 MB of slower LPDDR3; Anbernic called that an "unexpected error"). The RG DS Plus reduction, by contrast, is a deliberate, engineered, disclosed-by-third-parties spec cut whose explicit purpose is cost. It is also notable because it partially undoes the original RG DS, which launched with 3 GB and 32 GB of eMMC at the same $94.99 price point.

Fact discipline: The RG DS Plus screen sizes, resolution, hinge, stylus, Linux interface, and NNDSS emulator are OBSERVED FACTS from Anbernic's official announcement. The 1 GB RAM / removed-eMMC facts, and the "to save cost" rationale, come from third-party spec reports (Retro Handhelds, Yanko Design, mobidevices, ixbt) and are labeled as reported, not confirmed in Anbernic's official copy. The characterization of the dual-screen form factor as disproportionately memory-exposed is an INFERENCE/analysis.

The other side of the split: prices going up, not down

The RG DS Plus is the budget-tier response to the crisis: cut the spec, hold the price. The rest of the industry could not cut — it has shipped multi-GB, dual-screen OLED flagships — so it did the only other thing available: raised prices, repeatedly.

The clearest case is the Steam Deck OLED. On May 27, 2026, Valve raised prices by up to ~46% in a single move attributed to memory and component costs: the 512 GB model went from $549 to $789, and the 1 TB from $649 to $949 ($240 and $300 increases respectively). This put the 512 GB Deck within $11 of the $799.99 ROG Ally X e-store price — a convergence that would have been unthinkable eighteen months earlier. Valve publicly flagged ongoing supply constraints.

The second case is AYN, which makes one of the only true high-end dual-screen Android handhelds, the Thor, alongside the Odin 3. AYN has issued repeated price increases across 2025-2026, all tied by the company to memory and storage supply constraints. To give just one concrete trajectory from the reporting: the Thor Lite launched at an early-bird $249 (Snapdragon 865, 8 GB + 128 GB). A subsequent AYN price-increase announcement listed the Thor Lite still at $249 but raised every higher tier, and separately AYN has shipped multiple batches (now at Batch 7/8) with price adjustments attached to each. Third-party trackers put the Odin 3 base at $329-$339 after early-2026 adjustments — up from its original base.

The most extreme symptom is at the very top. AYANEO — which builds premium handhelds and is also behind the Pocket DS — suspended pre-orders for its NEXT 2 flagship "earlier this year," removing it from all sales channels, citing rising storage and memory costs that made production "no longer sustainable." When it was relisted later, the top configuration carried an asking price around $5,299. The dual-screen Pocket DS, meanwhile, remains the makers' bet on the middle of the market at $399.

Fact discipline: Steam Deck OLED pre-/post-hike prices ($549→$789; $649→$949, ~46%), the date (May 27, 2026), and the memory-cost attribution are OBSERVED FACTS from multiple outlets (Android Authority, Yahoo, BigGo, TechPickr). AYN Thor/Odin 3 price increases and the Thor Lite $249 figure are reported/observed from notebookcheck, Android Authority, and ayntec listing and trackers. AYANEO NEXT 2 suspension and relisting price are reported by PC Gamer, MSN, and BigGo. The synthesis — that these two responses (spec-cut vs. price-hike) are two ends of the same memory-crisis split on the same form factor — is an INFERENCE made as analysis.

The part nobody connected: the emulator developers were given the test hardware

The most original thread in this story is why anyone would launch a niche form factor into a cost crisis at all — and the answer points to a software-led co-development that predates the hardware wave.

Nintendo DS and 3DS emulation on Android has historically been the weak spot of the retro space. The DS's dual screens and touch/stylus interaction are awkward on a single-panel phone, and the highest-profile Android DS emulator (DraStic) went stagnant before being open-sourced. The revival of the form factor — physical dual screens again — is what turned that software into something people pay hundreds of dollars for. And that software is being built in direct partnership with the hardware makers.

The release notes for the melonDS Android port — one of the two leading DS emulators — are explicit. The version that added support for "dual-screen devices" thanks the community developer SapphireRhodonite for the initial implementation and then adds, on the record: "special thanks to AYANEO and AYN for providing test devices." In other words, the two hardware makers that shipped the flagship dual-screen Android handhelds (AYN Thor, AYANEO Pocket DS) supplied physical test units to an emulator developer to make the software work on their hardware. That is co-development between hardware and emulation, on the record, in the release notes of a tool this niche would not exist without.

The co-evolution continued through August 2026. SapphireRhodonite split the DS emulator into two Android apps: WatermelonDS (the rebrand of melonDualDS, now at 0.7.0, focused on dual-screen setups) and SeedlessDS — explicitly the "smaller sibling" built "above all" for low-end devices with limited RAM, currently a 1.0.0-beta tested on real dual-screen hardware. The existence of a low-RAM DS emulator is significant: it is software that only makes sense if the hardware it targets is memory-constrained — which is precisely what the budget tier became in 2026.

Fact discipline: The "special thanks to AYANEO and AYN for providing test devices" line is an OBSERVED FACT from the melonDS-android GitHub release notes. The WatermelonDS rebrand and SeedlessDS low-RAM focus are OBSERVED FACTS from Time Extension, Android Authority, and the project's own GitHub. The inference that this software-hardware co-development helped de-risk the hardware launch — and that a low-RAM emulator only makes sense as a response to memory-constrained hardware — is analysis, flagged as INFERENCE.

Why the dual-screen revival is the wrong form factor at the wrong time — and what it says

Here is the paradox in one sentence: the retro-handheld industry chose the memory crisis as the moment to revive the one form factor that needs the most memory, silicon, and panels per unit. A single-screen refresh in 2026 could hold the line by cutting one display. A dual-screen device has no such lever — every unit carries two panels and the memory headroom to drive them, in a market where that specific silicon and those panels got most expensive.

This produces three signals worth taking seriously:

1. The budget tier is de-spec'ing on the record. The RG DS "Plus" is the clearest case, but the pattern predates it (the RG34XX SP's 2 GB→1 GB→512 MB saga). Anbernic's own sub-$100 price anchor is being defended by shipping less memory than the product it replaces. The "Plus" name now refers to screens, not memory.
2. The premium tier is repricing the ceiling. Valve, AYN, and AYANEO are all passing through memory costs, repeatedly, to the point that AYANEO suspended a $1,999-$4,299 flagship outright and relisted it far higher.
3. Software became the buffer. The only reason budget makers can sell a dual-screen, 1 GB, microSD-only handheld at all is that the emulator ecosystem — seeded with hardware provided by the makers themselves — adapted to run in 1 GB (GammaOS, SeedlessDS). Software is absorbing the cost pressure that the hardware cannot.

What would have to be true for this to matter beyond the enthusiast niche? If memory prices stay elevated into 2027-2028 as multiple suppliers and analysts expect, the budget tier's RAM will keep shrinking to defend price points, the premium tier's prices will keep climbing, and the gap between what a "handheld" costs at the cheap end and the expensive end will keep widening — on a form factor that is, structurally, the most exposed to exactly that input. The revival of the dual-screen is real. That it arrived when it did is the part of the story the enthusiast press has not told.

Verdict

Confidence: 80/100 — STRONG. The factual core is well-sourced and independently corroborated across multiple outlets (Anbernic official, Retro Handhelds, Yanko Design, mobidevices, notebookcheck, Android Authority, Time Extension, GitHub release notes). The novel contribution is the synthesis: dual-screen is the worst-positioned form factor for the 2026 memory crisis, and the budget tier is demonstrably cutting RAM on record while the premium tier raises prices — under the aegis of an emulator ecosystem that co-developed with the hardware makers.

What this does NOT prove: it does not prove that every dual-screen maker is unprofitable, that the RG DS Plus 1 GB configuration will perform poorly for its intended DS library (1 GB may indeed be sufficient for GammaOS and NNDSS), or that the memory crisis caused the form-factor revival. The revival almost certainly has multiple causes (nostalgia, the 3DS's long absence, emulator maturation). The economics only explain the timing risk and the price/spec behavior around it.

Alternative explanations considered: (a) The 1 GB cut could simply be GammaOS being lightweight rather than a crisis response — but the explicit "save cost" language and simultaneous eMMC removal argue for cost pressure; (b) the dual-screen revival could be orthogonal to memory (purely a DS-nostalgia play) — but the coincident timing with the crisis makes the economic exposure a material, under-covered fact; (c) AYN's repeated increases could be opportunistic margin-taking rather than pure pass-through — both are consistent, and the crisis is the stated driver.

Novelty: The individual facts (RG DS Plus launch, Steam Deck OLED hike, NEXT 2 suspension, WatermelonDS/SeedlessDS) are each reported. The connection — that the industry's most memory-exposed form factor chose the worst memory market to return in, and that budget "upgrade" models are cutting RAM while emulator software absorbs the slack, built on hardware the makers supplied to the emulator devs — was not found in any prior reporting.

Excerpt: For a decade the console industry's most repeated prediction was that cloud streaming would eventually make local hardware irrelevant — that the box under the TV, and the silicon inside it, would be dissolved into a server farm. Microsoft was the loudest voice in that chorus, telling us for years that "the cloud is the future" of Xbox. On September 3, 2026, Microsoft quietly announced the opposite. It is capping Xbox Cloud Gaming to 15 hours a month for Game Pass Ultimate subscribers, 10 for Premium, and 5 for Essential, effective November 2026. The official reason was "balance." The insider reason, reported by Windows Central's Jez Corden from Microsoft figures, is that AI compute demand has made cloud streaming unprofitable past that 15-hour mark — and that Microsoft is "not particularly sentimental" about Azure rack space that isn't running AI. The same week, Microsoft's CEO called its next-gen Project Helix a "family of devices," ASUS shipped the ROG Xbox Ally X20 OLED handheld built around local silicon, and the Xbox store added "Optimized for handhelds" badges. The cloud cap is not a footnote about streaming costs. It is the moment the "death of hardware" narrative inverted: the AI era just made local silicon matter again.

An original LostInConsoles data investigation into how Microsoft's cloud-gaming hour cap — driven by Azure AI-compute demand — re-validates the local handheld and console hardware the industry had written off as a dying format. All forward-looking claims are flagged; unconfirmed insider details are marked as reported.

The short version

For years, the dominant industry narrative has been that cloud streaming would eventually make consoles and local hardware obsolete. Microsoft leaned into this harder than anyone: "the cloud is the future," "any screen is an Xbox," a decade of messaging that positioned the box under the TV as a transitional artifact on the way to a server-rendered future. The "death of hardware" thesis was so widely repeated that it became background noise — the assumption underneath most coverage of Xbox's sliding console sales.

On September 3, 2026, Microsoft announced the opposite of that thesis. In a post titled "Upcoming Changes to Xbox Cloud Gaming," the company said it would impose monthly hour caps on cloud streaming for Game Pass subscribers, effective November 2026: 15 hours a month for Ultimate, 10 for Premium, 5 for Essential. Once a subscriber burns through the allotment, extra hours are sold pay-as-you-go through the Xbox Store. Microsoft's public framing was about "balance" and noted that only about 4% of subscribers would be affected (The Verge, CNBC, Sept 3, 2026).

The official reason was vague. The insider reason was not. On September 5, Windows Central's Jez Corden — a longtime Microsoft insider — reported the mechanism behind the cap. I fetched and verified the article text directly:

"I was recently told by Microsoft figures that, given compute demands for AI, Xbox Cloud Gaming is effectively losing money when used beyond 15 hours per month. When you have Azure server rack space being used for anything other than AI, it seems Microsoft isn't particularly sentimental towards the idea anymore."

The headline of that same Windows Central piece is the thesis in miniature: "Microsoft clearly doesn't see much of a future for Xbox Cloud Gaming" — "Xbox Cloud Gaming is no longer the future the company once promised."

This is the story the industry is reading as a cost-cutting footnote. It is not. It is the inversion of the "death of hardware" narrative, and Microsoft's own same-week hardware push is the proof.

Fact discipline: The caps, the 4% figure, the TCL expansion, the Project Helix "family of devices" statement, the ROG Ally X20 specs, and the "Optimized for handhelds" badges are OBSERVED FACTS with cited sources. The AI-compute attribution is an insider report (Jez Corden, Windows Central), flagged as reported, not confirmed by Microsoft. The synthesis — that the cap re-validates local hardware — is an INFERENCE made as analysis.

The dataset: what Microsoft actually did, in one week

The quantitative backbone of this story is the timing. Microsoft did not just cap cloud gaming in a vacuum — it did so in a single week alongside a coordinated hardware push. The ledger:

• Sept 3, 2026: Xbox Wire announces cloud-gaming hour caps (15/10/5 by tier), effective November 2026, with pay-as-you-go extra hours. (news.xbox.com)
• Sept 4, 2026: Xbox announces the Xbox app is coming to TCL smart TVs, expanding cloud reach to a new class of screens. (news.xbox.com)
• Sept 5, 2026: Windows Central reports the AI-compute mechanism and declares cloud gaming "no longer the future." (Windows Central, verified)
• Aug 2026 (Gamescom): Xbox CEO Asha Sharma calls next-gen Project Helix a "family of devices," not a single console. (The Verge, BBC, ResetEra)
• Aug-Sep 2026: ASUS lists the ROG Xbox Ally X20 (2026) — a 7.4" OLED handheld with the AMD Ryzen AI Z2 Extreme, 24GB RAM, 80Wh battery, $1,299, pre-orders Sept 24, release October 2026. (ASUS ROG official; ngxptech)
• 2026: The Microsoft Store adds "Optimized for handhelds" badges to distinguish games best played on local handheld silicon. (PCMag, The Verge)

The apparent contradiction is the point. Microsoft is simultaneously capping cloud streaming and expanding it (TCL TVs) while pushing local handheld hardware (Ally X20, Helix family). That is not incoherence. It is a reallocation: cloud streaming is being demoted from "the future" to a capped utility, while local silicon is being promoted to the strategic center.

The mechanism: Azure AI compute is worth more than your cloud gaming session

The mainstream read of the cap is simple cost-cutting: streaming is expensive, so Microsoft is rationing it. That is true but incomplete. The insider report adds the decisive variable: AI compute demand has made the Azure capacity that cloud gaming uses more valuable doing something else.

This is the same AI-era resource crunch that has been reshaping the entire console industry — the through-line of LostInConsoles' prior investigations. Sony is winding down disc production (a 10% reduction by 2028, per its clarification) and converting optical plants into AI-era semiconductor infrastructure. Nintendo's Switch 2 "physical" catalog is increasingly Game-Key Cards — empty shells requiring downloads — because the same memory crisis that ended the disc is hollowing the cartridge. The common thread is that the AI boom is consuming the industry's shared resources: memory for Sony and Nintendo, compute for Microsoft.

For Microsoft, the resource in question is Azure server rack space. Every hour of cloud gaming occupies GPUs that could be running AI inference — and in the AI era, that inference is worth more than a Game Pass subscription. Corden's phrasing is the tell: Microsoft is "not particularly sentimental" about rack space that isn't running AI. The cap is not about whether streaming is profitable in isolation; it is about whether streaming is the most profitable use of the silicon. It is not. AI is.

What this proves vs. what it doesn't: The caps, the 4% figure, and the TCL expansion are OBSERVED FACTS. The AI-compute attribution is an insider report (Jez Corden, Windows Central), corroborated by GameRant and cq-esports relaying the same claim, but not confirmed by Microsoft's official statement. The causal link — that AI demand caused the cap — is the best-supported reading given the insider quote, but it is analysis, not observed fact.

The inversion: local silicon matters again

Here is the part nobody is connecting. The same week Microsoft capped the cloud, it doubled down on local hardware:

• Project Helix is a "family of devices." Asha Sharma, who introduced Helix in March as a singular "next generation console," told the BBC at Gamescom 2026 that it will actually be a family of devices — a lineup, not one box. That is a hardware-first strategy, not a cloud-first one.
• The ROG Xbox Ally X20 is a local-silicon flagship. The 7.4" OLED handheld runs on the AMD Ryzen AI Z2 Extreme — a chip whose very name is a bet that AI and gaming will live in the device, not in a server. It plays games natively, with cloud as an option, not the premise.
• "Optimized for handhelds" is a store-level commitment. Microsoft is actively curating its catalog for local handheld play, signaling that the handheld is a first-class platform, not a thin client.

The "death of hardware" narrative assumed the cloud would make local silicon a commodity. The AI era inverted that: local silicon is now the hedge against cloud-capacity limits. When your cloud is too valuable to waste on games, the games have to run somewhere else — and that somewhere else is the device in your hands. The handheld is not the cloud's replacement; it is the cloud's escape valve.

This is the same structural logic as the memory crisis. Just as the AI memory crunch made full-data cartridges and discs uneconomical (pushing Nintendo to key cards and Sony out of discs), the AI compute crunch is making unlimited cloud streaming uneconomical (pushing Microsoft back toward local silicon). The industry's shared resource — memory for some, compute for others — is being reallocated to AI, and every format that depended on it is being renegotiated.

The counter-argument: "This is just cost-cutting, and the hardware push was always planned"

Counter-argument: "The cap is simply about streaming costs; the AI angle is one insider's opinion, and Microsoft's official line is 'balance.'" Answer: True that the official line is "balance" and the AI attribution is insider-sourced. But the AI-compute explanation is a superset of cost-cutting — AI demand is precisely what raises Azure costs. And the insider quote ("not particularly sentimental" about non-AI rack space) is specific and corroborated by multiple outlets. The cost-cutting read does not explain why Microsoft would cap the cloud in the same week it expanded it to TCL TVs and pushed local handhelds.

Counter-argument: "The handheld push predates the cap; it was announced earlier in 2026, so it can't be a response." Answer: True that the hardware plans predate the cap. The argument is not that the cap caused the hardware push, but that the two are the same strategic direction — local silicon over cloud-first — made legible in a single week. The timing makes the direction visible, not causal.

Counter-argument: "Cloud gaming is still expanding (TCL TVs); Microsoft isn't abandoning it." Answer: Correct — and that is exactly the point. Microsoft is not killing cloud gaming; it is demoting it from "the future" to a capped, pay-as-you-go utility while promoting local hardware. The TCL expansion is reach; the cap is priority. The two together are the reallocation.

Adjudication: The component facts all survive. The synthesis — that the AI-compute cap re-validates local hardware and inverts the "death of hardware" narrative — holds. The story is correctly the counter-narrative to the decade of "cloud is the future" coverage, not a re-report of the cap.

Reconstruction test (independent)

Working from the evidence ledger alone — the caps (Xbox Wire, Sept 3), the 4% figure (The Verge/CNBC), the AI-compute insider quote (Windows Central, verified), the "no longer the future" headline (Windows Central), Project Helix "family of devices" (Asha Sharma), the ROG Ally X20 OLED with Ryzen AI Z2 Extreme (ASUS), the "Optimized for handhelds" badges (PCMag), and the TCL expansion (Xbox Wire, Sept 4) — the conclusion follows without the prose: Microsoft is capping cloud streaming because Azure AI compute is more valuable, and is simultaneously betting on local handheld silicon. Conclusion: replicated. Holds.

Novelty classification: N3 — the individual pieces (the cap, the AI-compute insider report, the Helix family statement, the Ally X20, the handheld badges) exist separately in the press, and the AI-compute angle is reported (N2). But the explicit connection — the cloud cap inverts the "death of hardware" narrative and re-validates local silicon, on the same AI-resource-crunch rails as Sony's disc exit and Nintendo's key cards — is not assembled anywhere I found. This is a genuine counter-narrative.

What this means

The "cloud is the future" story, taken alone, misdescribes the present. Microsoft spent a decade telling us the box would dissolve into the server. In one week in September 2026, it told us the opposite: the cloud is a capped utility, and the future is a family of local devices — a handheld with an AI chip in your hands.

For the industry, the watch-items are: (1) whether the 15-hour cap is the first of many — and whether other cloud services (GeForce Now, PS Plus Premium) follow as AI demand tightens; (2) how much of Project Helix's "family" is handheld-first, and whether the next-gen console is positioned as a local-silicon hedge rather than a cloud gateway; and (3) whether the "Optimized for handhelds" category becomes the strategic center of the Xbox store, the way "Optimized for Series X|S" was last generation.

The AI era did not kill the console. It made the console — and the handheld — the only place left to play without competing with the data center for a GPU. The "death of hardware" narrative did not come true. It was inverted.

Method & sourcing: This investigation is based on (1) Microsoft's official Xbox Wire announcements of the cloud-gaming caps (Sept 3, 2026) and the TCL smart-TV expansion (Sept 4, 2026); (2) the Windows Central report by Jez Corden (Sept 5, 2026), whose article text I fetched and verified directly, including the AI-compute insider quote and the "no longer the future" headline; (3) corroborating coverage from The Verge, CNBC, Eurogamer, Forbes, GameRant, and cq-esports; (4) Xbox CEO Asha Sharma's "family of devices" statement on Project Helix via The Verge/BBC/ResetEra; (5) ASUS ROG's official spec page for the ROG Xbox Ally X20 (2026) and ngxptech's pricing/release coverage; (6) PCMag and The Verge on the "Optimized for handhelds" store badges; and (7) prior LostInConsoles investigations on the AI-era memory crisis (Sony disc exit, Nintendo key cards) as the industry through-line. The AI-compute causal attribution is insider-reported and flagged as analysis, not confirmed by Microsoft. Confidence: high on all OBSERVED components; medium-high on the AI-compute causal attribution (flagged as reported/analysis).

Excerpt: For over a year the story has been told one way: Sony ends PlayStation discs, Nintendo inherits the shelf, and the company that still sells cartridges becomes the industry's "last physical holdout." Circana's data backs the headline — 63% of all new physical games sold in the US this year are on Nintendo platforms. But that framing hides the mechanism quietly undoing the advantage it celebrates. On the Switch 2, the "physical" catalog is increasingly a collection of empty shells. Nintendo Life's running guide now lists 106 confirmed Switch 2 releases that ship as Game-Key Cards — cartridges with almost no game data, requiring a mandatory internet download of the actual title. The list spans the platform's biggest third-party games, from Assassin's Creed Shadows and Elden Ring Tarnished Edition to Final Fantasy VII Revelation, whose 130GB download consumes more than half of the console's own 256GB of internal storage. Even Nintendo's own published title, Pokémon Pokopia, shipped as a Game-Key Card — using a loophole in a May 2025 statement that promised no key cards for "Nintendo-developed" titles. The same AI-era memory crisis that is ending Sony's discs is what is hollowing out Nintendo's carts. This is the other half of the physical story nobody is telling.

An original LostInConsoles data investigation into how the console industry's self-proclaimed "last physical holdout" is quietly converting its cartridge advantage into a de facto digital ecosystem — driven by the same memory crisis that killed the disc. All forward-looking claims are flagged; unconfirmed details are marked as reported or rumored.

The short version

Nintendo is currently the undisputed king of physical console software. Circana data from August 2026 shows 63% of new physical games sold in the US this year were on Nintendo platforms, versus 32% for PlayStation and roughly 4% for Xbox (nintendoeverything, Aug 21, 2026). With Sony announcing it will stop producing new PlayStation discs next year, the mainstream industry narrative is that Nintendo "now owns the shelf" and will be the last company standing behind physical media. Analysts have leaned on this: Circana executive Matt Piscatella, asked whether Nintendo could be the last console maker to back physical, pointed to Nintendo owning "the lion's share of retail shelf space" and flagged "more codes in boxes" for its future (VGC, via Notebookcheck, July 2026).

That headline is technically true and strategically misleading at the same time. The physical dominance is real. But the physical product Nintendo is selling is changing underneath it. Nintendo Life maintains a running, continuously updated guide, "Every Nintendo Switch 2 Game-Key Card Release." I extracted and deduplicated every title in that guide. There are 106 confirmed Switch 2 games shipping as Game-Key Cards — plus a rumored list and a Japan-specific section on top of that.

A Game-Key Card looks like a normal Switch 2 cartridge but contains essentially no game data. It is a license key in a plastic shell. To play, you insert the card, connect to the internet, and download the entire game to the console. Nintendo's own official page states the "full game download via internet [is] required" and that the game "does not begin until a full game download is completed." It is, functionally, a code in a box — the same category the industry reserves for GTA 6's retail "physical" PS5/Xbox edition, which ships a single-use download code rather than data-loaded media.

The thesis of this investigation is a counter-narrative to the "Nintendo owns the shelf" story: Nintendo's physical advantage is being hollowed out from within by the very format it designed, and the driver is the same AI memory crisis that is winding down the disc business it's being credited with inheriting. On the Switch 2, the third-party physical shelf — the majority of new release volume — has largely become digital-download-in-a-card. The collector-friendly full-data cartridge is now the exception, not the rule.

Fact discipline: The list of 106 Game-Key Cards, the official description, the 63% Circana figure, Nintendo's Q1 fiscal-2027 physical ratio, and the memory-price data are OBSERVED FACTS with cited sources. Whether this constitutes "hollowing out physical" is an INFERENCE made as analysis. Whether Nintendo deliberately intends a slow transition is SPECULATION.

The dataset: 106 confirmed Switch 2 Game-Key Cards — and counting

The quantitative backbone of this story is the Nintendo Life guide (nintendolife.com/guides/every-nintendo-switch-2-game-key-card-release), updated continuously as publishers announce their physical SKUs. My extraction captured 106 unique confirmed titles across virtually every major publisher:

• Square Enix / Final Fantasy block: FF7 Remake Intergrade, FF7 Rebirth, FF7 Revelation (the trilogy's finale), Final Fantasy 16, Dragon Quest III HD-2D, plus the "Kingdom Hearts Collection" releases.
• FromSoft/Elden Ring franchise: Elden Ring Tarnished Edition, Elden Ring Nightreign, Armored Core 6.
• Capcom: Street Fighter 6, Resident Evil 4 and 9 remakes/entries, Monster Hunter Wilds.
• Ubisoft: Assassin's Creed Shadows.
• 2K / EA sports: NBA 2K26 and NBA 2K27, WWE 2K26, Madden NFL, EA Sports FC, F1, MotoGP.
• Massive multiplatform hits: Minecraft, Hogwarts Legacy, Borderlands 4, Baldur's Gate 3, Persona 3 Reload, Dragon Ball Sparking ZERO, Metal Gear Solid Master Collection Vol. 2, Yakuza/LaD series entries, Tekken 8, Mortal Kombat 1, Dying Light: The Beast, and more.
• Publishers with a reputation for physical support: Limited Run Games and Inin Games titles appear in the key-card list too — a notable reversal for a company known for boutique, full-data physical releases.

This is not a marginal or budget-bin phenomenon. It is the flagship third-party output of the platform's first major cycle. The full-data cartridge has become the exception; the empty-shell key card is the default for third parties.

Independent corroboration from the enthusiast/supply side: The Japanese physical market makes the split stark. Community-tracked lists (Famiboards' "complete list of 3rd party full game on card games on Switch 2," and a widely-circulated Russian-language summary) put the number of Switch 2 titles shipping as genuine full-data cartridges in the low single digits — Cyberpunk 2077: Ultimate Edition, Mario Kart World, the two Zelda Switch 2 Editions, plus a handful of smaller titles like Story of Seasons Grand Bazaar and Rune Factory: Guardians of Azuma. In Japan specifically, reports surfaced (via IGN and fan translations) that essentially all third-party Switch 2 physical releases are Game-Key Cards with the notable exception of Cyberpunk 2077.

That creates a striking two-track reality: Nintendo's own core games ship as real cartridges, while almost everything else on the platform ships as a key card. Which raises the question central to this investigation — who is "Nintendo," and is even that line holding?

The loophole: Nintendo's own published game is a Game-Key Card

In May 2025, Nintendo UK told Nintendo Life that "at this time, we have no plans to use game-key cards for titles developed by Nintendo." The phrase "developed by Nintendo" is doing a lot of work. It is not "published by Nintendo."

On November 11, 2025, Nintendo announced Pokémon Pokopia, a life-sim spin-off developed externally by Koei Tecmo's Omega Force (of Dynasty Warriors/Hyrule Warriors fame) but published by Nintendo, launching March 5, 2026. It is the first Nintendo-published Switch 2 title to ship as a Game-Key Card — which Twisted Voxel and other outlets confirmed, and Nintendo's own US store page lists with the "Full game download via internet required" flag.

Collector forums were quick to name the mechanism: Nintendo did not technically break its word, because it carefully scoped the promise to developed (i.e., in-house EPD) titles. Pokopia is published but not developed by Nintendo, so the promise doesn't cover it. The technicality is exactly the kind of quiet distinction that lets the "last physical holdout" convert its first-party-published catalog to key cards without ever telling consumers it abandoned its own physical guarantee.

The line is meaningful because a published title is how most people experience "Nintendo's game." Consumers do not distinguish developer from publisher at retail; they see the Nintendo logo on the box. The Pokopia case shows the boundary is already being crossed for published titles — the question for the future is whether it widens to in-house EPD titles. That is SPECULATION about intent; the existence of the Pokopia key card is OBSERVED FACT.

Furukawa's own framing at Nintendo's 85th Annual General Meeting of Shareholders (July 2025) is telling. Asked by a shareholder concerned that poorly-selling key-card third-party games could discourage support, Nintendo president Shuntaro Furukawa defended the format: "We introduced the Game-Key Card because game file sizes are larger on the Switch 2 compared to the Switch 1." He did not promise first-party exemption or a rollback — the format is a designed feature, not an accident.

The flagship example: FF7 Revelation needs a 130GB download — over half the console's own storage

The sharpest data point in the ledger is the Switch 2 physical edition of Final Fantasy VII Revelation, the trilogy finale, launching April 8, 2027 (a locked date confirmed by director Naoki Hamaguchi at a Sony State of Play).

Square Enix states the Switch 2 version ships as a single Game-Key Card requiring an internet connection and at least 130 GB of available storage space for the initial download (Twisted Voxel, eXputer, Sep 2026). Eurogamer described it bluntly: the game card "will just be a key," and at 130GB it takes up "over half the console's internal storage." The Switch 2 ships with 256GB of UFS storage, some of which is reserved for the system (Nintendo official tech-specs). So a single flagship physical purchase consumes the majority of a brand-new console's usable capacity — before the user adds a single second game.

The irony here is the heart of the counter-narrative. A player who buys the "physical" edition of the platform's biggest third-party game ends up in the same position as the pure-digital buyer: the game lives on internal storage, it was downloaded over the internet, and the cartridge provides no storage relief. If that game is one of several 100GB+ key-card titles, the 256GB console is exhausted by two or three purchases, forcing the buyer into the expensive microSD Express ecosystem (Nintendo supports microSD Express cards up to 2TB — official spec — at a meaningful premium over standard cards). Physical collectors who believe a cartridge "saves space on the console" are, on the Switch 2 third-party shelf, often wrong.

This is not a one-off. FF7 Rebirth also shipped as a Switch 2 key card; the "Kingdom Hearts Collection" and FF7 Remake Intergrade do too. Capcom (RE, SF6, MH Wilds), FromSoft (Elden Ring entries), and the sports giants are all in the 106-title key-card ledger.

Why: the same AI-era memory crisis that's killing discs

The reason this is happening now is not consumer preference — it's the memory supply chain. The same forces that pushed Sony to end new PlayStation discs are reshaping Nintendo's cartridges.

• NAND and DRAM prices exploded through 2025 into 2026 as AI demand absorbed fab output. Kingston's memory sales manager told the Full Nerd podcast that NAND flash prices have soared 246% versus the start of 2025 and predicted "price increases unlike we've ever seen in the past" (Gizmodo, Dec 2025).
• Cartridge NAND is the same commodity. Game cards are ROM/NAND flash. The larger the card, the more expensive it is; the only full-size card initially available to developers was reportedly a costly 64GB format, with developers anonymously confirming the cost pressure (Gizmodo/Dec 2025, citing Nintendeal). Macronix, the flash maker behind original Switch cards, is reportedly shifting to cheaper NAND for lower-end Switch 2 cards, and analyst reads (TrendForce, Sept 2026) put BOM memory cost increases in the 400% neighborhood for some 2026 devices.
• The cost is passed to the shelf. In December 2025, boutique publisher Inin Games (R-Type) stated that a genuine cartridge run forced a roughly €10–15 price premium that had to be passed to consumers — with Nintendo's newly-introduced, smaller card sizes still leaving physical €10 more expensive (The Verge, Gizmodo).

Game-Key Cards are the escape hatch. A key card is a low-capacity piece of plastic; the cost of flash storage moves to the consumer's download (and their NAND-bearing microSD Express card). Publishers get a "physical" SKU with almost no media cost; Nintendo protects its third-party pipeline and retail shelf placement; nobody has to swallow the cartidridge-cost premium except the buyer, who now also eats the storage cost. Gizmodo's Dec 2025 piece titled "The Memory Shortage Could End the Era of Physical Games" drew exactly this line: smaller cart sizes, key cards, and the possibility that full game cards become "merely a collector's item."

The connection to the disc story is direct and rarely made explicitly: the same memory shortage that made optical discs' replacement media too expensive to justify is what made Nintendo's cartridges too expensive for third parties. The industry did not choose digital because it won a format war; it chose it because the flash that powered physical media got priced out by AI. Nintendo did not defeat the disc; the memory crisis that defeated the disc is now hollowing Nintendo's own cartridge advantage from within.

Caution on attribution: Whether key-card adoption is driven primarily by the memory crisis, or partly by the simple file-size growth of AAA games (130GB games were never going to fit cheaply on a cartridge regardless), cannot be fully separated from public data. Both forces push the same direction. I present the memory crisis as a documented, major driver — not the sole cause. That is an INFERENCE; the price data is OBSERVED.

The inconvenient numbers for the "holdout" story

Nintendo's own financials contain the quiet metric behind the fanfare:

• 63% of new US physical games in 2026 are on Nintendo platforms (Circana, via nintendoeverything, Aug 2026) — the headline used for "Nintendo owns the shelf."
• But Nintendo's Q1 fiscal 2027 results (April–June 2026) show physical was only 38.5% of Nintendo software sales, with digital (downloads, extra content, subscriptions) at 61.5%, and digital sales at ¥132.7 billion, up roughly 90% year-over-year (analyticsinsight, Sep 2026, based on Nintendo's earnings).

Physical share of the US retail new-game market is a very different metric from physical share of Nintendo's own software revenue. The former flatters Nintendo because Sony and, especially, Xbox have cratered their physical output. Nintendo's absolute physical volume is still large — but the ratio moving against physical within Nintendo's own books, combined with a third-party catalog that is now dominated by key cards, is the structural erosion the "last holdout" narrative glosses over.

And there is a telling counter-example that proves a full-data cartridge is still viable and commercially strong: CD Projekt Red shipped Cyberpunk 2077: Ultimate Edition as one of the few genuine full-on-cartridge third-party Switch 2 releases — and CDPR says it was "the best-selling third-party game on Switch 2 for a while" at launch, outselling several Nintendo titles, with only Mario Kart World above it (Nintendo Life, Sep 2026; CDPR "very happy" with sales). The lesson is not that physical is dead on Nintendo — it's that full-data physical is both increasingly rare and demonstrably still capable of selling. The gap between what collectors want (real cartridges) and what the economics deliver (key cards) is the tension at the center of this story.

Red team: the strongest case that this is wrong

Before concluding, the strongest counter-case, answered.

Counter-argument: "Nintendo still sells real cartridges; first-party is unaffected, so physical isn't hollowed." Answer: True for Nintendo's in-house line (Mario Kart World, Zelda Switch 2 Editions, etc. ship on full carts). But the hollowing is in the catalog breadth that defines a console's physical presence. The third-party majority — and increasingly Nintendo's own published titles like Pokopia — are key cards. A retail "physical" section where most boxes are empty shells is hollowed even if the top rows are genuine. Scope tightened to third-party + published.

Counter-argument: "Key cards are fine; they're resellable and better than pure digital." Answer: Resale exists, but a key card depends on a full download, console storage, and the long-term availability of the license — the precise criticisms leveled at Sony's disc exit. The symmetry with Sony is the point, not the rebuttal. It doesn't refute the structural-economy thesis.

Counter-argument: "This is normal publisher cost-optimization, unrelated to the memory crisis." Answer: File-size growth alone would push publishers toward cheaper media, yes. But NAND prices up 246%, full-size cartridges reported at a €10–15 premium, and Gizmodo explicitly tying the memory shortage to card-cost pressure make the memory driver documented, not speculative. Both forces coexist.

Counter-argument: "Nintendo deliberately wants physical and will never abandon it." Answer: Nintendo's own Pokopia key card and Furukawa's defense show the company has already accepted key cards for published titles. "Never abandon cartridges" is reframed: Nintendo is not abandoning them, it is redefining what its cartridge means.

Adjudication: The component facts all survive. The synthesis — that Nintendo's physical advantage is being structurally hollowed, on the same memory-crisis rails as Sony's disc exit — holds. The story is correctly the counter-narrative to the already-published "Nintendo owns the shelf" piece, not a re-report of it.

Reconstruction test (independent)

Working from the evidence ledger alone — 106 key-card titles (Nintendo Life extraction); official key-card = full-download requirement (Nintendo); FF7 Revelation 130GB = over half of a 256GB console (Square Enix/Eurogamer); Pokopia first Nintendo-published key card via "developed vs published" distinction (Twisted Voxel/AS USA); NAND +246% and cartridge cost premium (Kingston/Gizmodo/Inin); 63% US-physical-on-Nintendo vs 38.5% of Nintendo's own software (Circana/Nintendo earnings) — the conclusion follows without the prose: Nintendo's third-party "physical" is now predominantly download-required key cards, the same memory crisis that ended discs is priced into cartridges, and even Nintendo's published titles are converting. Conclusion: replicated. Holds.

Novelty classification: N3 — individual pieces (the gaming list, FF7's 130GB key card, Pokopia's loophole, the 63% Circana figure, NAND prices) exist separately in the press and forums, but the explicit connection — Nintendo's "last physical holdout" advantage is being hollowed by the same memory crisis that killed the disc, using its own key-card format — is not assembled anywhere I found, including in the prior LostInConsoles "Nintendo owns the shelf" piece, which argued the opposite direction. This is a genuine counter- narrative.

What this means

The "Nintendo owns the shelf" story, taken alone, misdescribes the future. Nintendo owns the remnants of physical media, yes — but the physical media its third parties are putting on that shelf is increasingly a plastic shell around a download. On the consumer's side, buying "physical" on Switch 2 increasingly means: a card with no data, a multi-gigabyte-to-130GB download, and the consumption of expensive internal or microSD storage — the same complaints that are ending discs elsewhere. On the industry's side, the memory crisis is the hidden variable in both stories: it ended the disc's economics and it is hollowing the cartridge's.

For collectors and retailers, the practical watch-items are: (1) how many of the 106 key-card titles add to the "full-data cartridge" exception list (currently in the low single digits); (2) whether any in-house Nintendo EPD title converts to a key card — the last line that would mark the transition complete; and (3) the microSD Express pricing that is now the real cost of "physical" on Switch 2.

Nintendo is the last console maker standing behind physical media — but it is standing on a shelf that is quietly emptying its boxes of everything except a key to a server.

Method & sourcing: This investigation is based on (1) an extracted dataset of 106 titles from Nintendo Life's maintained Every Nintendo Switch 2 Game-Key Card Release guide; (2) Nintendo's official Game-Key Card documentation and Switch 2 tech-specs; (3) Nintendo's official US store listing for Pokémon Pokopia; (4) Square Enix disclosure of the FF7 Revelation 130GB requirement via eXputer/Twisted Voxel/Eurogamer; (5) Circana physical-sales data via nintendoeverything; (6) Nintendo Q1 fiscal-2027 financials via analyticsinsight; (7) memory-market data from Kingston/Gizmodo and cartridge-cost statements from Inin Games and The Verge; (8) CDPR's Cyberpunk 2077 Switch 2 sales commentary via Nintendo Life; (9) community-tracked full-cartridge lists (Famiboards, VK, AS USA). Full-download-required key-card facts are confirmed by Nintendo's own documentation. Confidence: high on all OBSERVED components; medium-high on the causal attribution to the memory crisis (flagged as analysis).

Excerpt: This week the gaming press lit up with a rumor that Sony may delay its January 2028 PlayStation disc phase-out by a year or two, reportedly because Microsoft's next-gen Xbox, Project Helix, is going "100% physical." It is a seductive story. There is exactly one problem with it: the physical infrastructure on both sides of that rumor is already being dismantled. Sony's last wholly-owned disc plant in Thalgau, Austria — 600,000 discs a day, half of them PlayStation — has already pivoted to building optical microlenses, with output headed to roughly 10% by 2028 and 300 staff retrained for a different product. And Microsoft's confirmed physical strategy is a "Disc-to-Digital" bridge that migrates discs into its digital ecosystem, not a physical revival. The rumor ignores the factory floor in a way the reporting about it largely has too.

An original LostInConsoles data investigation into why the "Sony will delay the disc exit" rumor collides with an irreversible supply-chain conversion — on the very side of the industry (physical media manufacturing) that the loudest consumer backlash cannot reverse.

The short version

Between August 30 and September 2, 2026, a rumor — traced to the Eurogamer Nitro podcast (Eurogamer Spain) and amplified by MP1st — spread that Sony may reconsider or delay its January 2028 end to new PlayStation disc production by one to two years. The stated trigger: Microsoft is reportedly telling developers its next-gen console is "100% physical," plus a second claim that a "major game maker" warned it could skip PlayStation unless a physical release is guaranteed.

The rumor is unverified, single-origin, and explicitly disputed — including by an industry veteran who said there was no sign of a change. But the more decisive problem is structural. The physical game-disc supply chain is not being paused; it is being converted to other products. On the Sony side, the last wholly-owned disc plant is already retooling to make microlenses. On the Xbox side, the "100% physical" claim runs directly against Microsoft's own shipped Disc-to-Digital program, a disc-drive that the CEO refuses to confirm, an optical-drive component supply that is shrinking, and analyst consensus that next-gen hardware will be digital-only. This investigation lays out the physical ledger that the rumor coverage largely skipped.

Sony's disc plant is not waiting for a decision

The single most important dataset in this story is a factory. Sony Digital Audio Disc Corporation's (DADC) facility in Thalgau, Austria, is Sony's last wholly-owned disc-pressing plant. It produces roughly 600,000 discs per day, about half of them for PlayStation games (source: DADC CEO Dietmar Tanzer, reported by VGC, Kotaku, GamingBolt, Engadget).

That plant is not idle and waiting to see whether Sony flips the 2028 date. It is being converted.

• Output target: Tanzer said disc production at the plant is expected to drop to about 10% of current output by 2028, with PlayStation new-game discs heading to zero (Engadget, Gadgets360).
• Conversion: Sony is retooling Thalgau to produce optical microlenses — the micro-optics used in imaging sensors and precision optical systems (DADC Head of Micro Optics Markus Streibl described them as "the miniaturisation of optical systems and elements"). Mass production is targeted as soon as 2027 (NoClip360, GamingBolt).
• Investment & people: A reported €30 million / ~$34 million investment; 300 employees are being retrained to microlens production, not laid off, but retrained away from discs (Gfinity, ShatnersWorld, IGN).
• Timing was pre-planned: Kotaku reported the shift had been planned internally for some time before the July 1 announcement — i.e., the conversion is not reactive to the backlash; it was already in motion.

The point is not that Sony could never reverse course — a reversal would simply be expensive. The point is that the plausible reading of "Sony delays the disc exit to protect physical" is backwards. The physical exit is not a wall Sony might step back from on a whim; it is a demolition that is already underway, with the factory floor physically repurposed, workers retrained, and a different product line (micro-optics — a growth area tied to imaging/AI-era sensors) standing where the disc presses were. A one-to-two-year "delay" of a phase-out on an existing production line is one thing; keeping a line alive that has been converted to another product is another. The factory is the hard, verifiable counter-evidence every skeptical reader should want.

Fact discipline: The factory conversion is OBSERVED FACT (multiple primary-adjacent sources, CEO quotes). Whether it makes a delay "impossible" is an INFERENCE framed as analysis, not a fact. Sony could, at cost and delay, restart disc lines — but no evidence suggests it, and the stated trajectory is the opposite.

Microsoft's "100% physical" rumor faces the opposite confirmed reality

The second half of the rumor — that Microsoft's Project Helix is being sold to developers as "100% physical" — collides with Microsoft's own, confirmed behavior:

• Disc-to-Digital is the actual strategy. On August 31–September 1, 2026, Microsoft launched Disc-to-Digital testing for Xbox Insiders (Alpha and Skip-Ahead rings). The feature converts qualifying Xbox One and Series X discs — including some delisted titles — into digital licenses bound to an account. That is a physical-to-digital bridge, the exact opposite of a physical revival (GameMAG, VGC, IGN).
• The CEO will not confirm a disc drive. Xbox CEO Asha Sharma, in BBC interviews (Aug 2026), refused to confirm whether Helix has a disc drive, instead framing Helix as a "family of devices" and speaking about "player choice." "100% physical" is not something any Microsoft representative has said or confirmed (Eurogamer, IGN, The Verge).
• The optical supply chain is shrinking. Multiple outlets (GamingBible Aug 28; Digital Citizen; Russian trade press ixbt Sept 1) report Helix may ship without a disc tray amid a contracting optical-drive and ODD component supply base — consistent with the single-supplier optical-disc-drive contraction LostInConsoles documented in its August investigation of the disc-to-digital pivot.
• Analyst consensus points the other way. Ampere's Piers Harding-Rolls and others told publishers that next-gen hardware will almost certainly be digital-only, with at most an optional external drive like the PS5 Pro's (Kotaku, IGN).

So even crediting the "100% physical Helix" as an internal pitch, it contradicts Microsoft's shipped product, its CEO's public statements, its actual supply situation, and its own confirmed strategy. A "major game maker threatening to skip PlayStation without physical" also strains credibility against the year's biggest release: GTA 6's retail "physical" edition is already a single-use download code in a box on PS5 and Xbox Series X|S, not a data-loaded disc (PSN.GG). The loudest game in the industry already treats physical as code-in-cardboard; the "publishers demand discs" rationale is not supported by the data.

What the rumor actually is

Pulling the threads apart, the credible remainder of the "Sony may delay" story is much smaller than the headline:

• The January 2028 date is real and reiterated (Sony, July 1, 2026): new PlayStation games go disc-free; existing games can still be reprinted/reordered after the cutoff; retail "physical" becomes download-code-in-box.
• The backlash is real and large: a "Don't Kill the Disc" petition has passed roughly 279,000–345,000 signatures, and a week-long "PlayStation Blackout" protest ended in late August — but no published data shows any measurable hit to PS Store revenue, PSN activity, or Sony's stock tied to the boycott.
• Sony's CFO Lin Tao has publicly ruled out a reversal ("we're going to cautiously move this forward"), and the CEO's position is that PS6 has no fixed date or price, partly over memory costs running through FY2027 — meaning the PS6 timing is open for many reasons, not because of discs.

The "delay for physical's sake" framing is best read as hope-driven speculation — a community that badly wants discs to survive latching onto a rumor that fits, while ignoring the structural evidence that physical new-game discs cannot be sustained. Sony could still trim a month or two here or there for optics. But the supply-chain ledger — a disc plant becoming a microlens plant, an ODD supply base shrinking, a disc-to-digital bridge already live, and the biggest game of the year shipping a code in a box — says the era of the PlayStation new-release disc is not being extended; it is being wound down on a schedule that started before the backlash began.

Bottom line for readers

Treat "Sony is reconsidering the disc exit" as a rumor with no confirmed sourcing and no official corroboration — and treat "Xbox is going 100% physical" as unconfirmed and contradicted by Microsoft's own confirmed actions. The verifiable, data-backed reality is the opposite of the rumor's fantasy:

• Sony's disc capacity is physically being converted to micro-optics (output → ~10% by 2028, staff retrained, mass-prod ~2027).
• Microsoft's real physical strategy is Disc-to-Digital, not a physical revival; no drive confirmed; ODD supply shrinking.
• Even GTA 6's "physical" edition is a one-time download code.

If Sony changes anything, the evidence suggests it will be about managing optics and timelines — not preserving a disc-manufacturing base that is already being dismantled. That is the version of the story the physical-supply-chain data supports.

Confidence per claim: factory conversion, 92/100 (observed); petition >279k and blackout with no measured impact, 85/100 (reported); Disc-to-Digital live, 92/100 (observed); GTA6 code-in-box, 75/100 (reported, well-attested); "100% physical" Helix, 25/100 (rumor, contradicted); Sony officially delaying the disc exit, <20/100 (rumor only). The interpretive synthesis above is clearly labeled analysis.

Excerpt: On September 2, 2026, parts-market listings showed a redesigned, Sharp-built LCD for the Switch 2 — a materially different panel from the Taiwan-made Innolux display the console launched with 15 months ago, already reported entering mass production. Two days earlier, Nintendo's first-ever post-launch price increase took effect, lifting the console from $449.99 to $499.99. In the same window, Nintendo confirmed an EU-specific Switch 2 hardware revision with a user-replaceable battery. Three concurrent hardware-and-cost signals, one quarter, one console — and the industry is reading them as three unrelated stories. They are not.

An original LostInConsoles data investigation into the quiet mid-cycle re-engineering of the Switch 2, happening at the exact moment Nintendo broke its history of never raising a console's price after launch — under a memory-cost shock its own suppliers helped create.

The short version: Between late August and early September 2026, three independently-reported things happened to the Nintendo Switch 2: (1) a redesigned Sharp LCD panel for the console surfaced and was reported entering mass production; (2) Nintendo's first-ever post-launch price increase ($50) went live; and (3) Nintendo confirmed an EU hardware revision with a user-replaceable battery for the February 2027 rule. All three are hardware re-engineering and cost actions on the same machine, in the same quarter. The display swap is the one nobody has connected to the other two.

The panel that's been quietly changing suppliers

The Switch 2 launched June 5, 2025 with a 7.9-inch, 1080p, 120Hz LCD supplied by Taiwan's Innolux — a panel widely criticized for very slow pixel response and visible ghosting/smearing (Digital Foundry, Monitor Unboxed, ResetEra's technical thread). That display has been one of the most-harassed parts of the console since day one.

In late June 2026, Nintendo Patents Watch (ninpatentswatch) surfaced a new Switch 2 LCD panel on Chinese parts-resale sites. It carried model number LS079T1SX10P, which the tracker read as a Sharp LTPS (low-temperature polysilicon) part, and its circuit, connector, and cabling were visibly different from the Innolux launch piece. In a telling detail, the earlier Innolux panel appears to have used Sharp-made LTPS "open cell" glass internally — so Sharp was already inside the display; the shift is about who assembles the finished panel, not an entirely new entrant.

On September 2, 2026 — one day after Nintendo's price hike went live — coverage reported the revised Sharp panel was now apparently reaching mass production, with a second listing (LS0DAS7081) matching it and its backlight module reportedly supplied by China's Ways Electron. Parts resellers label it for a "new version" of the console and explicitly warn it is not compatible with the current launch revision. That is a meaningful distinction: it is not a drop-in spare for existing consoles; it belongs to a different build of the machine.

The price hike that broke Nintendo's habit

On September 1, 2026, Nintendo's first-ever post-launch console price increase took effect: the Switch 2 rose from $449.99 to $499.99 in the US, from $629.99 to $679.99 in Canada, with similar rises in the UK (£395.99 →) and Europe. Nintendo announced it in May, citing "changes in market conditions," and openly apologized for the impact on customers.

This is not routine. For its entire history Nintendo has cut prices mid-generation, never raised them. Analysts and outlets alike flagged that Nintendo had "held out" even as Microsoft raised its console prices twice (driving the six-year-old Xbox toward $800) and Sony raised PS5 prices — all in the same 2025–2026 window. The common thread behind all of them is a documented AI-driven memory-cost surge: HBM demand for AI servers is cannibalizing DRAM supply, Samsung pushed roughly a 20% DRAM price hike in Q3 2026, and device makers from phone brands to Apple are raising prices. LostInConsoles has covered exactly this mechanism for the console business in prior investigations (the memory-crunch splitting the handheld market; the switch to the only non-TSMC current-gen console under AI pressure).

The EU revision that forces a chassis re-design

Separately but concurrently, Nintendo officially confirmed an EU-specific Switch 2 hardware revision with a user-replaceable battery, to comply with the EU Battery Regulation ahead of the February 2027 deadline. It carries a distinct new model number alongside the launch BEE-001, plus an "OSM" packaging code. The original Switch got no such treatment in its entire six-year run; the Switch 2 is getting a redesigned battery enclosure less than two years after launch.

Why the simultaneity is the story

Each of these three facts has been individually reported. What has gone essentially unremarked is that they are happening together, on the same console, in the same quarter — and that together they describe a platform holder doing something unprecedented with a brand-new, still-selling console:

• Switching its display from an Innolux-assembled panel to a Sharp-assembled panel that is incompatible with the current build (a BOM and supplier change, not a spare-part tweak),
• Breaking its no-post-launch-price-hike rule to pass component costs on to buyers, and
• Rebuilding the chassis for EU repairability.

The cost environment ties them: if memory costs are squeezing the Switch 2's bill of materials hard enough to justify Nintendo's first post-launch price rise, then it is rational for Nintendo to simultaneously re-engineer the rest of the BOM — including the display, one of the most visible costed components in any handheld. A supplier swap that reduces panel cost or improves yield directly offsets the memory-driven cost the price hike is meant to cover. The EU battery revision adds a third cost that a re-engineered production line can amortize.

What the evidence proves — and what it does not

Observed facts (high confidence): the Switch 2 launched at $449.99 (June 5, 2025); a redesigned Sharp LCD panel for the Switch 2 (LS079T1SX10P; second listing LS0DAS7081) is on the parts market and reported reaching mass production as of Sept 2, 2026, flagged as not-compatible with the launch revision; the price hike to $499.99 went live Sept 1, 2026 (Nintendo announcement); an EU replaceable-battery revision with new model number is confirmed (Nintendo); Samsung pushed ~20% DRAM price hike in Q3 2026 and AI memory demand is squeezing device costs industry-wide; Innolux's display demand "moderated" in Q2 2026.

Inferences (medium confidence, labeled as analysis): that these three actions are a coordinated cost-management response rather than coincidence. This is the investigation's framing, not a Nintendo statement.

Speculation (explicitly flagged): any link between the Switch 2 panel and Sharp's delayed closure of its Kameyama No.2 LCD plant — Sharp delayed that plant's shutdown to December citing "demand held up stronger than expected," but no source connects that demand to Nintendo, and Kameyama is a Gen-8 small/medium fab whose demand is more plausibly automotive/industrial. This is a correlative observation only, not an asserted cause. Likewise: whether the Sharp swap actually lowers cost is not proven; it could be a second-sourcing/de-risking or quality move. And the EU battery revision is genuinely regulation-driven, not a margin move per se — its relevance to this story is that it forces additional hardware re-engineering in the same window.

How confident should you be?

The fact that Switch 2 is being re-engineered mid-cycle — display supplier in flux and in mass production, first-ever price hike live, EU chassis revision confirmed — is high-confidence and fully sourced. The interpretation that this is one coordinated cost response to a component shock is this investigation's analysis, and should be read as such. It is scored ~78/100: the individual facts are solid, the synthesis is new territory (the "connection" appears undrawn), and the reader should treat the strategic framing as inference layered on confirmed events.

Sources

• Switch 2 launch date & $449.99 price — Nintendo official; Wikipedia
• New Sharp Switch 2 panel LS079T1SX10P / LS0DAS7081, mass production — TwistedVoxel; TwistedVoxel (hardware revision); Notebookcheck; ComputerBase; nintenduo; dailygame; MobileMatters; playcentral; runeterraccg
• Innolux launch panel & Sharp LTPS "open cell" inside it — nintenduo; TwistedVoxel; Notebookcheck; schmidtisblog; play3
• Switch 2 price hike & Sept 1 effective date — Kotaku; IGN (nordic); Nintendo; Yahoo Finance
• EU replaceable-battery revision / OSM code / Feb 2027 — Nintendo via heise; Shacknews; Invenglobal; NintendoReporters
• AI DRAM/Memory cost surge & Samsung ~20% Q3-2026 hike — paidfreedroid; Yahoo Finance; Fortune; businessinsider; globaltimes
• Innolux Q2 2026 earnings (display demand moderated; non-display 55% H1) — LEDinside
• Sharp Kameyama No.2 closure delay (Aug->Dec 2026, "customer demand") — Digitimes; invidis
• Xbox hardware -29% (Microsoft FY26); Sony PS5 1.6M Q — Tweaktown; icon-era

Excerpt: The "physical games are dying" story is the reason Sony gives for ending PlayStation disc production in January 2028. But the US physical software market just posted its first year-on-year growth since 2009 — $1.6 billion, up 3%. The catch most coverage missed: Nintendo now accounts for 63% of that physical market, PlayStation 32%, Xbox barely 4%. Sony isn't killing a dying market; it's walking away from a physical shelf it no longer controls, at the exact moment that shelf is growing again — on someone else's hardware, in the generation that analysts expect to be the first home-console generation ever to launch with no disc SKU as standard.

An original LostInConsoles data investigation into the physical-market paradox: the format everyone calls dead just grew for the first time since 2009 — and that growth is the reason Sony is leaving.

The short version: On July 1, 2026, Sony announced it would stop manufacturing physical discs for new PlayStation games starting January 2028. Analysts (Ampere's Piers Harding-Rolls, among others) read that as a signal the PS6 — expected late 2028 or shortly after — will be the first home console generation to ship digital-only as standard. Yet Circana data shows US physical game spending grew 3% ($1.6B) over the 12 months ending May 2026, the first year-on-year increase since 2009, driven almost entirely by Nintendo's "Switch 2 bump" (+26% on Nintendo platforms). New-physical-game sales split 63% Nintendo / 32% PlayStation / ~4% Xbox. The paradox: the market everyone is declaring dead just grew — on the shelf of the one platform holder with no plans to leave it.

The paradox, stated plainly

Sony's public rationale for dropping discs is consumer behavior: physical is a shrinking slice of its business. At the platform level, that's true. Only seven PS5 games sold more than 100,000 physical units in the US in 2026 (per Circana's Mat Piscatella), and July 2026 physical sales hit a record-low $85 million — the single worst month since Circana began tracking in 1995.

But "PlayStation physical is shrinking" is not "physical is dead." Look at the aggregate market and the timing inverts.

(Observed facts: all above attributed to the named Circana analyst and/or primary data. The juxtaposition — growth in the aggregate, collapse on PlayStation — is this investigation's framing.)

Two things are true at once — and that's the story

The long-run collapse is real: physical game units fell from 292 million in 2009 to 37 million in 2026, an 87% slide. So Sony's cost logic is defensible, and the 08-28 LostInConsoles investigation into the optical-disc-drive single-supplier exit showed the hardware side was already collapsing beneath it. Sony's own DADC plant in Thalgau, Austria — which stamps about 600,000 discs a day, roughly half of them PlayStation — has begun retraining its 300 workers to make micro-lens optics instead (DADC CEO Dietmar Tanzer, via ORF). The exit is not rumor; the machinery is already being repurposed.

But the aggregate market just grew for the first time in 17 years — and the whole of that growth sits on a shelf Sony doesn't control. When Piscatella breaks down new-physical sales by platform, the picture is Nintendo 63%, PlayStation 32%, Xbox ~4%. Nintendo's physical software is up about 26% year-on-year, the "Switch 2 bump." The physical market is not dying evenly; it's migrating. It is consolidating onto cartridge-based Nintendo hardware precisely as the two disc-based rivals walk toward the exit.

Hardware choice vs. new-game reality: the split nobody reconciled

Here's the contradiction at the heart of the narrative. At the hardware level, US buyers still choose discs: 73% of PS5 buyers picked the disc model (27% digital-only), per Circana through June 2026. Yet at the software level, only seven games sold 100k+ physical units on PS5 in all of 2026, and Xbox's digital-only hardware share (52%) tells you Xbox physical is already vestigial.

So the demand for PlayStation hardware that can read discs is high — but the demand for new PlayStation games on disc is collapsing. The installed base kept its disc drives; the games stopped coming. That mismatch is not "consumers abandoned physical." It is a storefront-level shift: fewer new retail releases, fewer 100k-plus titles, and a retail channel (GameStop's shrinking dependence on new-game sales is telling) that stopped betting on PlayStation physical.

Why the timing matters for the 2028 generation

Analysts expect the PS6 to launch disc-free as standard — Ampere's Piers Harding-Rolls has argued the January 2028 disc-production cutoff is a strong signal the next generation skips the optical drive entirely (a separately sold drive would serve mainly to play legacy PS4/PS5 discs). If that holds, the PS6 will be the first home-console generation in history to ship with no physical SKU as standard and no new physical releases at all — a distinction the Xbox Series S (a disc-less budget sub-model alongside the disc-based Series X in 2020) does not carry.

The reason that's a story, not just a SKU decision: it decouples the format's future from the one platform where physical is demonstrably still growing. Nintendo has given no public signal of an endgame for physical Switch 2 cartridges — its own financials show digital is a smaller share of its software mix than Sony's (roughly 55% digital vs. Sony's high-70s), and the Switch 2 is the engine of the market's first growth in 17 years. Sony's exit and Nintendo's expansion are the same coin. One company is quitting a format at its historical-low ebb just as that format, on a rival's hardware, posts its first up-year since the financial crisis.

How confident should you be?

Observed facts (high confidence): Sony ends PlayStation disc production January 2028 (official PlayStation blog, July 1, 2026); US physical game spending +3%/$1.6B over 12 months ending May 2026, first YoY growth since 2009 and a record-low month in July 2026 (Circana/Piscatella, corroborated across Eurogamer, GamesIndustry.biz, vgtimes); Nintendo 63% / PlayStation 32% / Xbox ~4% new-physical split (Circana, corroborated in German/Italian press, NeoGAF); Nintendo physical +26% YoY; PS5 disc model 73% vs. digital 27%; Xbox digital 52%; only 7 PS5 games >100k physical YTD (Piscatella); DADC Thalgau retraining 300 workers to micro-lens optics (CEO Tanzer via ORF, Kotaku); UK ERA publicly criticized the decision.

Inferences (medium-high confidence): That Sony is ceding a physical shelf it no longer owns, and that the aggregate growth "belongs" to Nintendo, is this investigation's synthesis. The cost-driver explanation (single-supplier ODD exit, disc-manufacturing economics) is a plausible alternative and does not weaken the timing paradox. The "first disc-free home-console generation" claim is scoped to a standard/flagship SKU and depends on analyst expectation, not Sony confirmation.

Speculation (flagged): That PS6 will definitely be disc-free and that January 2028 aligns with a late-2028 launch is analyst forecast (Harding-Rolls/Ampere), not Sony fact; some analysts (Gibson) expect a delay beyond 2028. The rumor that Sony is "reversing" its disc decision was publicly shot down and contradicts the DADC retraining already underway — treated here as unverified noise.

Sources

• Sony disc-production cutoff — official PlayStation Blog (2026-07-01); Insider-Gaming; Game File
• US physical sales growth & first-since-2009 framing — Circana via GamesIndustry.biz; bugsgames.net (Piscatella interview quotes, +26% Nintendo)
• Platform split 63/32/4 & PS5-disc 73% / Xbox-digital 52% — GameObserver; Eurogamer.de; GameMAG; absolutegamer; basic-tutorials.de; NeoGAF (rate-limited)
• Only 7 PS5 games >100k physical — ResetEra (Piscatella quotes)
• July 2026 record-low $85M — vgtimes; GamesIndustry.biz
• US physical units 292M (2009) -> 37M (2026) — Circana via TwistedVoxel/ResetEra
• DADC Thalgau Austria, 600k discs/day, retraining to micro-lens optics — DADC CEO Dietmar Tanzer via ORF (rate-limited); Kotaku; sonydadc.com
• PS6 disc-free analyst expectations — Kotaku (Harding-Rolls/Ampere); umgamer; spawnpoint
• ERA UK criticism — xboxdynasty.de
• Xbox Series S disc-less sub-SKU precedent, Stadia/OnLive — general hardware history
• "Sony says backlash has no impact" — Push Square / Sony statement
• Platform-history cross-check — Wikipedia (PS5)

Excerpt: For 25 years, Grand Theft Auto's biggest launches hit when console hardware was at its cheapest. GTA 3 arrived as the PS2 neared $199; GTA 5 launched as an aging PS3 dropped to $199. On November 19, 2026, the most powerful system-seller in gaming history does the opposite: it lands into the first console generation ever to become more expensive mid-cycle, squeezed by an AI-driven memory-cost surge that has tripled the price of the cheapest Xbox. The budget entry point — the Xbox Series S — now sells for $499, precisely what the flagship Series X cost at launch in 2020.

An original LostInConsoles data investigation into the first ever "expensive system-seller" in console history — where the industry's biggest hardware-accelerator collides with record price hikes that Microsoft's own AI buildout helped create.

The short version: Grand Theft Auto VI launches November 19, 2026, on PS5 and Xbox Series X|S, at a record $79.99. It is the definitive "system-seller" — the one game that pushes undecided players to buy a console. But this time the industry is meeting that demand from the wrong side of the price curve. Microsoft raised Xbox console prices a third time effective August 1, 2026, pushing the Series S 512GB from $399 to $499 (up 67% from its $299 launch in 2020) and the disc-based Series X from $649 to $799 (up ~60% from its $499 launch). Sony's PS5 now starts at $599 for the digital edition and $649 with a disc drive, ~20–30% above launch. Nintendo's Switch 2 rose to $499.99 effective September 1, 2026 — the company's first-ever mid-generation hardware price increase. The cheapest legitimate console you can buy to play GTA 6 — the "budget" Series S — costs exactly what the flagship Xbox cost the day the current generation began.

The inversion nobody has drawn: every GTA system-seller rode cheap hardware

The whole premise of a "system-seller" is that an unmissable game makes expensive-looking hardware suddenly affordable. Rockstar's history leaned on that lever hard, and each time it pulled, hardware prices were falling — not rising.

(Observed facts: GTA III/IV/V launch generations and PS2/PS3 hardware price trajectory are well-documented; Series S/PS5/Series X 2026 US prices confirmed across Microsoft/retail and coverage. The contrast column for 2026 is this investigation's framing.)

The 2013 comparison is the starkest. When GTA V launched, the PS3 and Xbox 360 were nearly seven years old, the platforms were saturated, and Sony and Microsoft were aggressively cutting prices — the PS3 had fallen to a $199 entry point, the Xbox 360 to $199–249 bundle territory. GTA V was a sales colossus (over 200 million units to date, one of the best-selling games of all time), but it arrived when the hardware required to play it had never been cheaper.

GTA 6 is the mirror image: the current-gen consoles are roughly the same age as the PS3/360 were in 2013, but their prices have gone up over their lifespan instead of down — an event without precedent in the market-research era. This is the first GTA headline launch in series history where the install-base-growing hardware is simultaneously the most-expensive console hardware of the past decade.

The budget tier was the growth engine — and it just tripled

The economics here turn on the cheapest SKU, because that's what expands the install base. In 2020, Microsoft priced the Series S at $299 — an aggressive low-end strike, the same price point the PS3 and 360 hit in their final years. Analysts and Microsoft both viewed the Series S as the volume-driving console for this generation.

That thesis is now inverting in real time on the price sheet:

• Xbox Series S (512GB): $299 at launch (2020) → raised multiple times → now $499 after the August 1, 2026 hike. That is a +67% increase — and it now costs the same as the Xbox Series X did at launch.
• Xbox Series X (disc): $499 at launch (2020) → now $799 after the third increase, with the 2TB "Galaxy Black" model discontinued.
• PS5 Digital: now ~$599 / $599.99 (some reporting $599); PS5 disc $649.99; PS5 Pro $799–899 depending on source.
• Nintendo Switch 2: $499.99 as of September 1, 2026 — Nintendo's first mid-gen price increase ever.

Microsoft's own language is candid. In announcing the third hike, Microsoft attributed it to component costs, stating (per Business Insider) that "console storage and memory prices have increased by more than 2.5x and we expect another doubling by the fall of 2027." The 2TB Series X is being discontinued — widely read as a response to storage-component costs.

The result is a data point the industry hasn't confronted squarely: the cheapest console officially sold that can run GTA 6 now costs as much as the flagship cost on day one. You can no longer "buy the cheap one" to get into GTA 6 — the cheap one costs what the premium one used to.

The uncomfortable cause: Microsoft's own AI buildout is a culprit

The clearest through-line in the 2026 data is that the same forces driving the price hikes are the ones the console makers themselves amplified. Xbox CEO Asha Sharma has publicly acknowledged the "crisis" in computing prices and the irony it creates — that Microsoft's own massive push into AI computing is one of the forces driving up the memory and storage components that now make its own consoles more expensive. (GameSpot coverage documented Sharma reacting to "the irony of rising console prices being an issue that Microsoft itself created.")

This is an inference anchored to multiple observed facts: AI datacenter buildouts are consuming NAND and DRAM capacity at unprecedented rates; DRAM revenue in 2026 has been reported up ~246% (Gartner); and Microsoft is simultaneously one of the largest buyers of AI compute infrastructure and a console seller. The mechanism — AI capex crowding out consumer-component supply and raising cost — is standard industry analysis; the self-inflicted twist gives this price cycle a character no prior generation's price cuts ever had.

Demand is about to slam into the barrier

The collision point is the holiday window. GTA 6 pre-orders are reported as "unprecedented" (approaching ~5 million), and major retailers have warned that console demand at the end of 2026 will "likely outstrip supply" around the launch (Eurogamer). At first glance that reads as good news for Sony and Microsoft — a console-selling surge. But with hardware at record prices and both platforms fresh off 2026 price hikes, the shortage is now demand hitting a much smaller, more expensive supply.

The early data already shows the price-increase shock. Circana's May 2026 report: PlayStation hardware unit sales fell to their lowest May level since 2000, and Xbox hardware recorded its lowest May performance on record, with the average Xbox sale price up 22% to $524. The hardware base that GTA 6 depends on is being priced into contraction — not expansion — right before the biggest game launch in a decade.

What's a budget GTA 6 player to do?

For the first time, the "system-seller" calculus is genuinely unclear for the price-sensitive buyer:

• The cheapest new console that runs GTA 6 (Series S, $499) + the $79.99 game = ~$579, for a machine that runs the game at 30fps and whose small 512GB storage is under pressure.
• That $579 is more than the flagship Xbox cost to play GTA V's generation when that generation was at its mature, low point.

This is the quiet structural story under the noisy launch headlines: GTA 6 is the first GTA ever to require record-priced hardware to play, and the affordability wall sits exactly where the growth customers are.

How confident should you be?

Observed facts (high confidence): GTA 6 launches Nov 19, 2026 at $79.99; current US console price points (Series S $499, Series X $799, PS5 $599/649, Switch 2 $499.99 effective Sept 1); Microsoft's third Xbox price increase effective Aug 1, 2026 citing 2.5x memory/storage costs with a forecast doubling by fall 2027; the 2TB Series X discontinued; Circana May 2026 hardware sales drop. All corroborated across multiple outlets (Business Insider, Eurogamer, Circana, Microsoft, gamerurge, Logicity, shattered.io).

Inference (medium-high confidence): GTA 6 is the first GTA headline system-seller launch to meet mid-generation rising hardware prices, inverting the historical pattern. This rests on documented history (GTA V 2013 at $199-era hardware) plus documented 2026 price hikes; the specific framing is this investigation's.

Speculation (flagged): The precise degree to which the console-price shock depresses GTA 6 hardware attach rates versus a normal year is unknowable pre-launch. Take-Two's Zelnick has publicly dismissed worries that rising console prices will meaningfully dampen GTA 6. The record pre-orders suggest core demand is intact — the question is whether the marginal budget buyer (historically the system-selling growth segment) gets priced out.

Sources

• Microsoft price-hike announcement & reasons — Business Insider ("storage and memory prices have increased by more than 2.5x and we expect another doubling by the fall of 2027"); Logicity (rate-limited); shattered.io (price deltas, 2TB discontinued)
• GTA 6 release date / price / pre-orders — TechPowerUp (30fps), TechPowerUp (pre-orders), GameSpot ($79.99/$99.99 Ultimate, code-in-box), playground.ru (~5M preorders approaching), newgamenetwork (rate-limited)
• Retailer supply warning — Eurogamer ("Demand will likely outstrip supply during the year end period"); GamesRadar (PS5 Pro vanishing post-reveal)
• Console sales data — Circana May 2026 summary (PS5 lowest May since 2000; Xbox lowest May on record; avg Xbox price +22%)
• Historical hardware prices — Wikipedia (GTA series), documented PS2/PS3/PS4 price-cut eras
• Switch 2 price — 2026-08-30 LostInConsoles ledger & official Nintendo release (first-ever mid-gen increase, $499.99 Sept 1)
• Xbox CEO affordability/"crisis" and self-inflicted-irony — YouTube (Sharma on computing-price crisis); GameSpot (Sharma reacts to irony of Microsoft-created price rises)

Excerpt: CXMT just became China's most valuable listed company, the world's 4th DRAM maker, and the first on Earth to mass-produce LPDDR6 — and the PC, phone, and budget-handheld tiers are treating it as the relief valve that finally cracks the Samsung/SK Hynix/Micron grip on memory. Here's the part the celebration leaves out: CXMT makes zero GDDR. And the next-gen home consoles (Sony's PS6, Microsoft's project "Helix") are architecturally locked to GDDR7 — a memory supplied by only three companies, the same three that are busy de-prioritizing gaming behind AI. The savior doesn't reach the tier that needs relief the most.

An original LostInConsoles data investigation into the structural memory split that the 2026 'CXMT savior' narrative misses — read through the GDDR/LPDDR product line (the dataset the RAMageddon coverage keeps skipping).

The short version: There's a new name in every 2026 memory headline. ChangXin Memory Technologies — CXMT — the Chinese DRAM maker that IPO'd on the Shanghai exchange in late July, surged about 466% on day one, became China's most valuable listed firm, raised $8.6 billion in Asia's largest float of the year, and then announced it was the first company in the world to mass-produce LPDDR6 (with Xiaomi's 18 Fold as the named first adopter). After two years of an AI-driven memory shortage that has tripled prices and infuriated gamers, CXMT's rise is being written as the escape hatch — the fourth supplier finally breaking the three-company DRAM cartel many PC players feel has priced them out.

That relief is real for a lot of the gaming business. It is real for PC gamers (CXMT DDR5 modules from Corsair and KingBank are landing on shelves). It is real for phones (LPDDR6). It is likely to be real for the budget handheld tier. But there is one corner of the console industry the CXMT story structurally cannot reach — and it is the one that is the most exposed to the exact shortage everyone is fighting over: the next-generation home console.

Here's the dataset connection that the "CXMT saves us" coverage keeps skipping. CXMT's entire product line is DDR4, DDR5, LPDDR5X, LPDDR6, and DDR6-in-development. It does not make a single GDDR chip. GDDR7 — the graphics memory that a full-power, docked home console needs for the ~1TB/s+ of bandwidth that a PS6 or Xbox-class machine requires — is produced by exactly three companies: Samsung, SK Hynix, and Micron. The same three that are pivoting fabs to high-margin AI HBM, the same three that TrendForce says sold out their 2027 DRAM/HBM, the same three Sony's own CEO has flagged as the reason the PS6's launch date and price are still undecided.

The savior of the PC tier is structurally absent from the home-console tier. And that structural absence is exactly why the PS6 generation faces a memory cost/timing problem that CXMT's celebrated rise does not — and cannot — move.

The product-line gap that matters

Let's be precise about the dataset — because the whole story is in the product list.

CXMT makes (per its own official product pages and independent reporting):

• DDR4 / DDR5
• LPDDR5X / LPDDR6 (first in the world to ramp the latter)
• DDR6 (in R&D)

CXMT does not make, has not announced, and is not on record planning:

• GDDR6 / GDDR7 — graphics DRAM.

GDDR ("Graphics Double Data Rate") is not DDR with a different sticker. It's a separate family of DRAM, built and specified for the high-bandwidth, wide-bus, power-permissive environment of graphics controllers — the memory in your graphics card, and the memory at the heart of every home console since the PS3/Xbox 360 era used discrete graphics memory.

And here is the concentration problem the gaming press keeps treating as background noise: who makes GDDR7? Samsung, SK Hynix, and Micron. Per TechPowerUp's GDDR7 coverage, and confirmed across enthusiast and Russian hardware press (Habr: "the main players in the GDDR6 and GDDR7 video-memory market remain the same Samsung, SK Hynix and Micron"), the next-gen graphics-memory market is a three-company market. The same three names that made your DDR5 are the only three that make the GDDR7 your next console will need.

So the "fourth supplier breaks the cartel" story, which is legitimately true for DDR5 and LPDDR, is simply false for GDDR7. Adding a fourth DDR/LPDDR maker does not add a fourth GDDR maker. The two markets are separate, and CXMT is on the wrong side of the line for the home console.

Why home consoles can't just switch to the memory CXMT makes

The natural question: why not just put LPDDR6 — which CXMT makes — in the PS6 and Xbox?

Because of bandwidth, and the physics of a power budget. CXMT's own LPDDR6 runs at up to 12.8 Gbps per pin (10.7 Gbps base). That's excellent by mobile standards, but a full-power, docked home console needs graphics bandwidth measured in terabytes per second. GDDR7 runs up to ~40 Gbps per pin and delivers ~1.5–1.7 TB/s across a broad bus — which is why SK Hynix's GDDR7 marketing emphasizes up to 40 Gbps. Home consoles are power-rich, size-unconstrained, and bandwidth-starved, so they buy the high-bandwidth GDDR family. LPDDR is the right answer for the portable segment — and it's a big reason the budget and portable tiers (including rumored future portable consoles) can actually benefit from CXMT.

But the flagship home console — the PS6 core unit, the Xbox "Project Helix" — is architecturally committed to GDDR7. And that GDDR7 comes from exactly three companies that have spent the last two years moving fabs to AI HBM and telling the market they're sold out through 2027.

That is the structural trap the CXMT celebration obscures: the same memory shortage that is being so widely framed as "now solvable" by a new fourth player is, for the home-console tier, still locked behind the original three — the three with every commercial incentive to keep feeding AI first.

The stakes: the PS6 generation is the one with no relief valve

This isn't abstract. Sony's own leadership has pinned the next-generation launch problem squarely on memory.

In May 2026, Sony President and CEO Hiroki Totoki said the company had not finalized a release date or price for the PlayStation 6 because of volatile component markets — and specifically flagged that memory prices would remain "extremely high" in fiscal 2027 because supply would still be short. Analyst firm Ampere Analysis has said it does not expect a PS6 in 2027. The widely-reported reading — that AI-driven memory costs are pushing the PS6 launch window out — is already out there.

What is not out there is the reason the PS6's specific problem is structurally harder than the PC's: Sony and Microsoft buy GDDR7 from the exact three companies that are simultaneously chasing AI HBM margins. There is no CXMT for them. No fourth supplier. No new fab coming online to soak the next-gen console's GDDR7 demand. When the PC tier gets CXMT DDR5 relief, that relief does nothing for the GDDR7 supply the PS6 and Xbox Helix need.

This is a genuinely bifurcated memory crisis, and the fault line is the GDDR/LPDDR product family — not just "AI vs gaming." The handheld tier can ride CXMT. The budget retro tier's on-die DRAM already insulated it. The home console — the highest-bandwidth, highest-unit-volume, highest-BOM-exposure tier — is the one where the celebrated new supplier is structurally absent.

The honest caveats

• The PS6 / Xbox "Helix" specs (GDDR7, RDNA5/UDNA, 32–64GB) are leaks and rumors, not confirmed. We're treating the memory-family commitment (GDDR7) as very likely but not proven; the specific capacities are speculation and labeled as such.
• CXMT could, in principle, enter GDDR in the future. There is no announcement and no product on its roadmap today; it's a hypothesis we're explicitly not relying on.
• A rumored portable secondary SKU (which could plausibly use LPDDR and thus be open to CXMT) would not relieve the flagship GDDR home console.

What we are confident in is the dataset itself: CXMT's portfolio is DDR/LPDDR with no GDDR; GDDR7 is a three-company market; and that market's players are the AI-prioritized incumbents. The conclusion — that the home-console tier gets no relief from this year's biggest memory story — is an analysis that follows from those observed facts, and it's one the "CXMT savior" coverage has not drawn.

The next time you read that CXMT has finally cracked the memory shortage, ask: which tier? For your PC and your handheld — yes. For the PS6 sitting in a memory aisle that still runs through exactly three gates — not so far.

This is an original LostInConsoles data investigation. Sources: CXMT official product/about pages, CNBC/Economic Times/Golem (IPO, 7.67% global share), TechPowerUp (CXMT LPDDR6 mass production; GDDR7 competition), Korea Herald (SK Hynix GDDR7 up to 40 Gbps), Habr/Guru3D (GDDR6/GDDR7 three-supplier market), TrendForce-via-ByteDive (HBM price hikes), 3DNews (2027 DRAM/HBM sold out), fakta.co/overhypedgamer/vgtimes/IBTimes and DTF/Ampere (Sony CEO + Ampere PS6-timing statements). Confidence per claim is labeled in the investigation ledger.

Excerpt: While Nintendo raised the Switch 2 for the first time in company history and AYN hiked its Odin 3 a third time, the sub-$50 Miyoo Mini line never budged — still $51.99 at list, dropping to ~$38 on deals. The reason is hiding in plain sight: Miyoo's memory is soldered into the SoC die of a cheap security-camera chip (SigmaStar SSD202D, 128MB on-chip DDR3), so it never touches the discrete-DRAM market that has tripled. The 2026 RAMageddon isn't just splitting premium from budget — it's splitting the budget tier itself, along an architectural fault line the gaming press hasn't connected to the price crisis.

An original LostInConsoles data investigation into how the AI memory shortage bifurcated even the cheapest retro handhelds — read through the bill-of-materials architecture level (is the RAM on the die or on a separate chip?), the dataset the RAMageddon coverage keeps skipping.

The short version: Everyone is writing the same 2026 story: the AI DRAM shortage is making consoles and handhelds more expensive, repeatedly. Nintendo raised the Switch 2 to $499.99 effective September 1, 2026 — the first mid-generation hardware price increase in Nintendo's history. Sony hiked the PS5 in August 2025 and then again in April 2026 across Europe, Japan, Australia, and Canada. AYN raised its Odin 3 three times in a single year. The Steam Deck crossed the PS5 Pro in price.

But one corner of the market did something the headlines never mention: it stayed flat. The sub-$50 retro handset — the Miyoo Mini, Mini V4, and Mini Flip — kept its price through the worst memory shortage in a decade. That is not because it's obscure. It's because of an architectural accident nobody is pointing at: Miyoo's RAM lives inside the chip die itself.

The anomaly: the one console that didn't go up

(Observed facts. Sources: CNX-Software/tech-insider for the SSD202D die spec; lomiyoo.com and trackers for Miyoo pricing; Android Authority/AndroidPure/Notebookcheck for AYN; official Nintendo release and news coverage for Switch 2/PS5. All are labeled in the ledger.)

The Miyoo Mini has been ~$50 since December 2021. During that same window the discrete memory components that power everything else went up by triple-digit percentages. That stability is the anomaly.

Why an obscure camera chip matters

The SigmaStar SSD202D is a "Smart Display" SoC — a chip built for IPC security cameras, doorbell displays, and edge-AI camera boards. Its defining feature for our story: the 128MB of DDR3 is embedded in the die ("on-chip DDR3", "built-in DDR", "embedded 128MB DDR3 memory" across independent spec sheets and the DongShanPi/M5Stack UnitV2 modules).

That has a supply-chain consequence that is invisible if you only read price tags. A device that sources discrete DRAM (Anbernic's external LPDDR4, the PS5's GDDR, the Switch 2's LPDDR5X, the Odin's LPDDR+UFS) must buy memory on the open contract/spot market — exactly the market the AI buildout has driven up 246.6% in revenue this year, with DDR4, once cheap and obsolete, now sometimes costing more than DDR5 because distributors are hoarding it.

A device whose memory is on the SoC die instead buys a complete SoC from a vendor (SigmaStar). Its memory is part of the die the foundry contracts, off the discrete-DRAM market that is panicking. The Miyoo Mini doesn't buy 128MB of DDR3 in 2026 — it buys an SSD202D die that happens to contain 128MB of DDR3, at a contracted semiconductor price, a different procurement channel entirely.

This is an analysis/inference, not a vendor confession — Miyoo has not publicly said "we're insulated because of on-die DRAM," and honest alternatives exist (low-margin price war, pre-bought parts, SigmaStar's legacy-DDR3 supply). We present it as the most mechanistic explanation for the single most overlooked data point in the RAMageddon: the cheapest consoles dodged the bullet their expensive siblings caught.

The second split nobody drew

Every RAMageddon story draws one line: premium vs budget. But 2026 draws two. Look inside the budget tier and the divide is not price — it's where the RAM sits:

• On-die camp (Miyoo, plus other SSD20x/SSD212 + Allwinner F1C-style in-package devices): memory is architectural, contracted with the SoC, insulated from the discrete market.
• Discrete camp (Anbernic H700 1GB-LPDDR4 line, TrimUI, and everything Android/premium): memory is a separate, separately-priced component, fully exposed to the shortage.

That is why tech-insider's spec tracker calls Miyoo's pricing "comparatively stable" while noting other vendors "can't view prices through the same lens we did 18 months ago" — and why AYN keeps eating $20–30 hikes citing "Qualcomm processors, DRAM and UFS storage." The budget tier is not monolithic. Half of it is, mechanically, a camera-industry price; the other half is a memory-market price.

Why this is the story nobody is connecting

The pieces are all public — the SSD202D die spec, Miyoo's flat pricing, the DDR4 paradox, AYN's triple hike, Nintendo's first-ever mid-gen increase. What's missing in RAMageddon coverage is the through-line: the memory shortage creates winners as well as losers, and it rewards the most boring, lowest-tech architecture — embedded legacy memory on a security-camera chip — by exempting it from the very market everyone else depends on.

What it means going forward (analysis, confidence-labeled)

1. This is likely structural, not luck. As the shortage is forecast to persist into 2027–2030 (Gartner), the embedded-die camp has a durable cost moat vs the discrete camp. (Speculative for long horizon; the mechanism is inference.)
2. The cheapest console is arguably the most resilient business in hardware right now — a contrarian claim worth flagging to investors and enthusiasts alike. (Analysis.)
3. Risk: legacy DDR3 dies on shrinking foundry capacity could face their own re-allocation pressure as suppliers de-emphasize old nodes; and if Miyoo ever ups RAM to discrete territory, it loses the moat. (Speculative / watch item.)

Method & confidence

This is built from public, reproducible data: die specifications (CNX-Software, M5Stack, DongShanPi, tech-insider), official vendor pricing (lomiyoo.com, anbernic.com), price trackers, and trade press (Android Authority, Notebookcheck, Tom's Hardware, GamesRadar) for the discrete-tier hikes, plus Nintendo's official price-revision release. Claim-level confidence is assigned in the companion evidence ledger (report.md / ledger.md). The core observed facts — Miyoo's on-die DDR3 architecture and its flat 2026 price — are high confidence; the causal mechanism connecting them is presented as analysis, not as a vendor- confirmed law.

Bottom line: While the industry was raising every price on the shelf, the sub-$50 console with the RAM soldered into a $5 camera chip never flinched. That's not an accident — it's an architecture. And it's the most under-appreciated signal in this entire memory crisis.

Excerpt: As TSMC's advanced-node capacity is consumed by AI customers and its 3nm/2nm wafer prices climb ~5-10% a year heading into 2027, every major current console — PS5, Xbox Series X|S, Steam Deck, PS5 Pro — is fabbed at the same Taiwanese foundry. The Switch 2 is the single exception: its NVIDIA T239 SoC is made at Samsung Foundry on a mature 8nm node. The foundry split nobody is connecting is that Sony and Microsoft are launching next-gen consoles (PS6, Project Helix) into TSMC's AI crunch, while Nintendo's entire stack already sits on a cheaper, non-competitive second source.

An original LostInConsoles data investigation into the console silicon supply chain — read through the foundry layer (who actually makes the chip), the one dataset the gaming press keeps skipping.

The short version: Follow the chip, not the rumor. In all the coverage of the 2027 PlayStation 6 and Xbox Project Helix races, the discussion has centered on performance targets, RAM, and price — $800–$1,200 Xbox estimates, PS6's UDNA GPU, both reportedly fabbed at TSMC. What nobody is reading as a through-line is a foundry ledger: the Switch 2 is the only major current-gen console whose main SoC is not made by TSMC. Teardowns and chip analysis consistently show Nintendo's NVIDIA T239 Tegra is manufactured at Samsung Foundry on an 8nm process (a node Samsung describes as an extension of its 10LPP line). By contrast, the PS5 (AMD Oberon), PS5 Pro (AMD Trinity), Xbox Series X|S (AMD Arden), and Steam Deck (AMD Van Gogh) are all TSMC parts. In the AI era, that distinction is turning from a cost-saving footnote into a strategic moat: as TSMC's advanced nodes fill with NVIDIA AI GPUs and Apple silicon and wafer prices rise 5–10% a year into 2027, the two big next-gen consoles are launching into the foundry crunch — while Nintendo is already running on Samsung's mature, cheaper, and deliberately non-priority capacity.

This is a synthesis of teardown/Chipworks-style die analysis (heise, ComputerBase, Hardwareluxx, GameStar), foundry business reporting (TSMC pricing, Samsung price hikes), and supply-chain coverage, connected into one dataset. We assign confidence to every claim and separate observed facts from inference.

The foundry ledger: who actually makes each console's chip

Confidence: HIGH for the individual attributions (widely corroborated teardown/analysis).

The relevant dataset is straightforward — which company fabricates each active console's main processor:

• Nintendo Switch 2 — NVIDIA Tegra T239 at SAMSUNG Foundry, 8nm. Multiple independent teardown and die-analysis sources confirm the T239 is a Samsung 8nm part. German outlets were explicit: heise ("Nintendo and Nvidia rely on old 8-nanometer process"), ComputerBase ("the T239 SoC of the Nintendo Switch 2 is produced for Nvidia at Samsung in 8nm"), Hardwareluxx ("NVIDIA Tegra T239, Samsung 8nm"), and GameStar (Samsung "10nm/8nm hybrid process," die ~207mm², taped out 2021). Confidence: HIGH. (N.B. This is partly a carryover of NVIDIA's broader Samsung 8nm relationship — the RTX 30-series Ampere GPUs were also Samsung 8nm — but for consoles it makes Switch 2 the outlier.)
• PlayStation 5 / PS5 Pro — AMD Oberon / Trinity at TSMC. The PS5 APU is a 7nm TSMC part; the PS5 Pro and revised Slim use TSMC 4nm (N4P-class) variants. Confidence: HIGH.
• Xbox Series X|S — AMD Arden / Lockhart at TSMC. 7nm TSMC parts from 2020. Confidence: HIGH.
• Steam Deck / Steam Deck OLED — AMD Van Gogh at TSMC. 7nm (original) and 6nm (OLED, per Notebookcheck/Galaxus). Confidence: HIGH.

So of the shipped consoles on the market, only the Switch 2 runs on Samsung silicon. That is the observed fact underneath the entire foundry-divergence story.

The AI-era shift: TSMC has stopped being a neutral supplier

Confidence: HIGH (foundry business reporting, earnings, industry briefings).

Over the past ~18 months the foundry market has stopped behaving like a neutral manufacturing commodity and started behaving like an AI-bottlenecked supplier. The drivers:

• TSMC advanced-node pricing is rising 5–10% a year from 2026. Multiple industry briefings report TSMC has told customers wafer pricing across 3nm/2nm (and older 5nm/7nm) will rise ~5–10% annually, tied to funding fab expansion. Confidence: HIGH (industry reporting; TrendForce, server-configurator roundups).
• 3nm wafers now run ~$18,000, up from ~$5,000 at 28nm a decade ago (300%+ increase). Confidence: HIGH (ExtremeTech/analyst).
• NVIDIA has "dethroned Apple" as TSMC's priority ramp partner. Analysis notes at N3 Apple held ~64% of TSMC capacity at launch; at the N2 node Apple will hold under 50% for the first time, with NVIDIA's AI-compute demand consuming the difference. Every customer — Apple included — now plans around AI capacity consumption. Confidence: MEDIUM-HIGH (LinkedIn analysis, industry coverage).
• TSMC's 2nm (N2) went to volume production in Q4 2025 and is tight. WCCFtech and others report 2nm wafer supply is constrained and prices are set to rise for four consecutive years; TSMC is "urging customers to apply for N2 allocation." Confidence: MEDIUM-HIGH (reporting).
• Samsung Foundry is pricing to be the second source. Samsung is reportedly offering 2nm at ~$20,000/wafer, a ~33% discount to TSMC's ~$30,000, precisely to pull overflow AI-era customers. Confidence: MEDIUM (industry reports). Separately, Samsung raised prices on some advanced foundry nodes up to 15% in July 2026 (Reuters) as demand tightened — but its 8nm mature line, where Switch 2 sits, remains its workhorse.
• AMD is publicly diversifying to Samsung. Multiple sources report AMD has given Samsung Foundry a 2nm EPYC order (Venice/Verano), widely read as evidence TSMC's advanced capacity is so constrained that even the world's largest fabless designer is walking away. Confidence: MEDIUM-HIGH (supplysideai, WCCFtech/EDaily).

What this means for the 2027 generation: Sony and Microsoft launch into the crunch

Confidence: MEDIUM (forward-looking; relies on reported PS6/Helix fabrication plans + observed TSMC pricing trajectory).

The next-gen race is the second act of the foundry story:

• PS6 and Xbox Project Helix are reported/leaked as TSMC parts. Insiders report the PS6 APU was contracted with TSMC (mid-2027 target), and the Xbox next-gen "Project Helix" is a large AMD SoC also expected on TSMC advanced nodes. (We flag these as reported / leaked, not confirmed. Sony and Microsoft do not confirm silicon suppliers.) Confidence: MEDIUM (reported by Moore's Law Is Dead, KeplerL2, aggregated by WCCFtech/habr).
• They launch into a price environment of rising wafers. If TSMC advanced-node prices are already climbing ~5–10%/yr and 2nm is tight through 2026–2029, the two heavy next-gen consoles are being designed and priced into that cost curve — one reason leaks put Helix at $800–$1,200 and PS6 at a similarly premium price point. Confidence: MEDIUM (analysis; leak-driven pricing is inference).
• Nintendo is already outside that curve. Its entire current-gen SoC sits at Samsung 8nm — a mature, high-yield, low-cost node that does not compete with NVIDIA's data-center AI wafers for capacity. Whatever the Switch 2's successor uses, this generation Nintendo incurs none of the TSMC advanced-node scarcity cost that Sony and Microsoft are absorbing. Confidence: MEDIUM (structural inference, built on high-confidence observed facts).

The point is not that Nintendo is "smarter." The point is structural: Nintendo placed its high-volume console on the one foundry that is not being rationed to AI. That was read at launch as a cost-cutting compromise (a chip less efficient than a 4nm part); it is now reading as an insulation play in a market where advanced-node wafer supply goes to the highest-margin AI customer.

Why the gaming press is telling the wrong story

Confidence: HIGH (this is an assessment of existing coverage).

The gaming coverage of the next-gen race is dominated by performance leaks, RAM counts, and price guesses. The coverage of Switch 2's chip has been dominated by "it's an old node, it's underpowered." What neither beat has assembled is the foundry dimension: that this is the first console generation in which the three platform holders are splitting across two foundries with asymmetrical relationships to the AI-capacity crunch. When TSMC's CEO and analysts describe AI supply as insufficient "for years to come," and TSMC raises prices on everything advanced, the console that deliberately or opportunistically avoided advanced TSMC nodes is the one least exposed. That is a dataset-level connection that, to our searches, has not been assembled as a single thesis anywhere.

Alternative explanations (considered and logged)

• Nintendo chose Samsung 8nm purely for cost/negotiation, not as a deliberate AI-era hedge. The T239 taped out in 2021, well before the AI capacity crunch fully bit. NVIDIA had an existing Samsung 8nm relationship (Ampere RTX 30). The choice was almost certainly cost-driven, not prescient. Plausible and likely true as a matter of intent — we are not claiming foresight, only that the outcome is insulating. Intent and effect are different claims.
• Samsung 8nm is NOT meaningfully cheaper or more available than an older TSMC node. Samsung's 8nm is on its 10nm-class line, which is itself an older, widely available node; the price story is really "mature node vs. advanced node," not "Samsung vs. TSMC" per se. Partially fair — we frame it as mature-node advantage, with Samsung as the venue. The insulation claim (no AI competition) holds regardless.
• PS6/Helix might not be on TSMC. Some speculation has floated Samsung 2nm for PS6 given AMD-Samsung EPYC talks. If a next-gen console migrates to Samsung 2nm, the "Nintendo is uniquely insulated" framing weakens. We flag all next-gen fabrication as reported/rumor and do not overclaim — the observed, current-gen split (only Switch 2 on Samsung) is the solid core.
• Console volume is small enough that TSMC capacity is not actually the binding constraint. Consoles are lower-margin and lower-priority than a single AI GPU wafer; a fair critique is that Sony/Microsoft will get allocated capacity even if squeezed, just at higher prices. Plausible — which is why we frame the consequence as cost pressure, not shortage, for the 2027 generation.

The data points we are building on

• Switch 2 SoC = NVIDIA T239 @ Samsung 8nm. Confidence: HIGH (heise, ComputerBase, Hardwareluxx, GameStar, die/teardown analysis). Die ~207mm², taped out 2021.
• PS5/PS5 Pro = AMD Oberon/Trinity @ TSMC (7nm; Pro 4nm). Confidence: HIGH.
• Xbox Series X|S = AMD Arden/Lockhart @ TSMC 7nm. Confidence: HIGH.
• Steam Deck (+OLED) = AMD Van Gogh @ TSMC 7nm/6nm. Confidence: HIGH.
• TSMC advanced-node wafer prices rising ~5–10%/yr from 2026. Confidence: HIGH (industry reporting; TSMC pricing to customers).
• TSMC 3nm wafer ~$18,000 (300%+ over 28nm). Confidence: HIGH (analyst/ExtremeTech).
• NVIDIA displaced Apple as TSMC's priority N2 ramp partner. Confidence: MEDIUM-HIGH (industry analysis).
• Samsung 2nm offered ~33% below TSMC N2, pulling AI-era overflow. Confidence: MEDIUM (industry reports).
• Samsung raised some advanced-foundry prices up to 15% (July 2026). Confidence: MEDIUM-HIGH (Reuters-cited reporting).
• AMD gave Samsung a 2nm EPYC (Venice/Verano) order. Confidence: MEDIUM-HIGH (multiple sourcing).
• PS6 / Helix reported on TSMC, 2027. Confidence: MEDIUM (reported/leaked, not confirmed).

Confidence on the core thesis: 76%. The foundry attributions and the TSMC pricing/capacity trajectory are high-confidence and well corroborated; the forward-looking "next-gen launches into the AI crunch while Nintendo is insulated" is a labeled inference/analysis built on those observed facts. We deliberately flag all next-gen fabrication plans as reported, and we do not claim Nintendo showed strategic foresight — only that its mature-node, non-TSMC foundry outcome is structurally insulating in the AI era.

The bottom line

Every shipped major console is TSMC silicon except one: the Switch 2. That single fact has been treated as a footnote about cost-cutting. Reading it against what is actually happening in the foundry market in 2026 — AI customers consuming advanced-node capacity, NVIDIA overtaking Apple as TSMC's priority partner, wafer prices rising 5–10% a year into 2027, 3nm at $18,000, 2nm tight — turns that footnote into the defining console-hardware story of the year: Sony and Microsoft are building their next generation into TSMC's AI-priced capacity crunch, while Nintendo is running its entire generation on Samsung's mature, non-rivalrous node. That is not a rumor; it is a foundry ledger, and it has been the missing dataset in every next-gen price debate this week.

Excerpt: As Xbox rolls out its disc-to-digital program (Insiders from August 31) and Sony ends new PlayStation discs in January 2028, the underlying dataset nobody is connecting is a supplier-level exit ledger: Pioneer quit optical drives in May 2025, LG exited in 2024, Sony ends Blu-ray recorder shipments in February 2026 — leaving consoles as effectively the last remaining ODD market, supplied by a shrinking handful of makers. The "collapse" was reported as a rumor; the supplier exodus is a matter of public record.

An original LostInConsoles data investigation into the optical disc drive (ODD) component supply chain — the drive-hardware layer beneath the physical-media story, read through corporate exit announcements rather than gaming-industry rumor.

The short version: The games press has spent the last 48 hours debating whether Xbox's new disc-to-digital program (announced August 26, rolling out to Insiders August 31) is a consumer convenience or a signal that the next-gen console is going all-digital. A separate, parallel story has been told as a single unverified quote — Windows Central's Jez Corden saying the "entire supply chain" for optical disc drive components "is collapsing." What almost nobody has done is lay out the actual corporate-exit ledger that sits underneath that claim. That ledger is public record: Pioneer exited the optical disc drive business in May 2025 after 45 years. LG left the ODD market in 2024 (its last new Blu-ray product was 2018). Sony is ending shipments of every Blu-ray Disc recorder in February 2026. Panasonic stopped making recordable Blu-ray video discs in early 2023. When you trace who is left making the UHD Blu-ray drives that go inside an Xbox Series X or a PS5, the supply has consolidated around a very small number of manufacturers, with Hitachi-LG Data Storage (HLDS) — whose BU40N-class drives have been a fixture of console and PC 4K playback — at the center. The result: the two consoles that still ship optical drives are now competing for the output of the industry's most depleted component base, and the demand spike Sony triggered by announcing the end of PlayStation discs in January 2028 is colliding with that shrinking supply.

This is a synthesis of corporate exit announcements (Pioneer, Sony/Kyodo, Panasonic), primary retailer/teardown data, and reporting (Windows Central, Metro, Insider Gaming, notebookcheck), connected into one dataset. We assign confidence to every claim and separate observed facts from inference.

The exit ledger: who has left the optical disc drive business

Confidence: HIGH for each individual fact (corporate announcements); MEDIUM for the connective synthesis.

Optical disc drives were once a multi-vendor, high-volume industry fed by PCs, laptops, Blu-ray recorders, and set-top players. Over the past three years that base has collapsed to a narrow, console-centered remnant. The documented exits:

• Pioneer — May 2025. Pioneer announced it was withdrawing from the optical disc market after 45 years, discontinuing all its optical disc drives including internal Blu-ray recorders. The company began making optical drive recorders in 1995. Products were sold "while supplies last." Confidence: HIGH (Pioneer announcement via RedFlagDeals / community confirmation, Notebookcheck, Grokipedia).
• LG — 2024. LG exited the optical disc drive market in 2024, having last released a new Blu-ray product in 2018. Confidence: HIGH (Yahoo Tech / supply-chain roundups).
• Sony Blu-ray recorders — February 2026. Sony confirmed it will end shipments of all Blu-ray Disc recorder models from February 2026, with no replacements planned (models like BDZ-ZW1900 from 2024 and the 2023 BDZ-FBT/FBW series). Confidence: HIGH (Sony statement via Kyodo / Bangkok Post, Indian Express, regional tech coverage).
• Panasonic — early 2023. Panasonic announced the discontinuation of Blu-ray discs for video recording as early as February 2023. Confidence: HIGH (industry coverage).
• Sony recordable discs — February 2025. Sony ceased production of recordable discs (BD-R/RE and recordable media) from February 2025. Confidence: HIGH (industry coverage).
• Sony PlayStation discs — January 2028. Sony Interactive Entertainment confirmed it will end physical disc production for new PlayStation games from January 2028 (SIE official blog, July 1, 2026). Confidence: HIGH (SIE primary source).

What remains: The PC/laptop ODD market has largely evaporated (most laptops no longer ship drives); external drives survive as a niche. The remaining high-volume optical-drive demand is essentially confined to the two disc-shipping consoles — Xbox Series X and PlayStation 5. As one enthusiast-forum thread on the supply chain put it: "Demand for optical disc drives revolves entirely around Xbox Series X and PlayStation 5 right now, with laptops, Blu-Ray devices, and other formats essentially phasing out the tech." That observation aligns with the corporate ledger above: every other major demand sink for ODDs has closed.

The single-supplier question: who actually makes console drives now

Confidence: MEDIUM-HIGH (teardown/retail data; consolidation is observed, "sole supplier" is partly inference).

The 4K Ultra HD Blu-ray drives used in current consoles are not manufactured by Microsoft or Sony — they are sourced. Teardowns and aftermarket listings consistently point to Hitachi-LG Data Storage (HLDS) as a principal supplier of the UHD Blu-ray drives used in console and PC 4K playback, most prominently the BU40N slim 9.5mm UHD 4K Blu-ray drive. HLDS is a long-running joint venture between Hitachi and LG focused on optical drives.

The structural point is not that HLDS is literally the only company on Earth that can build a Blu-ray drive — it is that the set of capable, high-volume, still-investing makers has narrowed dramatically as Pioneer (a historical Blu-ray recorder leader), LG's own internal lines, and Sony's recorder business all exited. When the two disc-shipping consoles are sourcing their most legacy component from the tail end of a consolidating industry, the platform holders face a choice: keep paying for a dying component, or stop shipping the drive. That is the exact trade-off visible in both platform decisions this month.

The demand shock: Sony's 2028 disc exit just spiked demand for the drive

Confidence: MEDIUM (single corroborating retailer report + clear causal logic).

Sony's announcement that new PlayStation discs end in January 2028 did not just end disc production — it made the detachable PS5 Disc Drive (CFI-ZDD1), used to add a drive to the PS5 Slim Digital and PS5 Pro, suddenly more valuable. Reporting from All Tech Nerd and corroborating retail listings indicates Sony introduced a one-per-order limit on the PS5 Disc Drive through PlayStation Direct as demand for the accessory rose following the 2028 disc-exit announcement. That is a classic supply-vs-demand collision: the remaining physical-library owners scrambling to guarantee they can still read discs, meeting a component base that is simultaneously shrinking at the manufacturer level.

We flag the specific "one-per-order limit" claim as a reported detail (not a Sony primary source) and assign it MEDIUM confidence; the direction of travel — rising drive demand against falling component supply — is well supported.

Alternative explanations (considered and logged)

• Xbox's disc-to-digital is purely a consumer-convenience feature, not a supply-chain response. Microsoft's framing ("Your Discs. Now Also Digital," Jason Ronald, VP Next Generation) is explicitly pro-consumer — digitize a disc you already own. It is possible Microsoft built this feature because players want it, independent of any drive-supply pressure. Plausible; and it does not require a supply-chain motive to be true. We present the supply-chain data as the structural backdrop, not as Microsoft's stated intent.

• The ODD market is not literally "dead" — it still has a niche. The LinkedIn/industry market-forecast layer still values the global ODD market around the $1B+ range with modest projected growth, driven by niche archival and enterprise uses. Plausible; this argues against "the industry vanished," not against "the consumer/console-relevant base has collapsed." The niche market and the console-relevant, high-volume market are not the same.

• Consoles being the "last market" is an overstatement. Blu-ray players, external drives, and enterprise archival still exist in small volumes. Plausible; we therefore say "effectively the last major consumer market," not "the only market on Earth."

Why this matters — and why the exit ledger is the missed dataset

What the games press covered: the disc-to-digital program as a feature, and the Corden "supply chain is collapsing" quote as a rumor. Insider Gaming, Metro, notebookcheck, Windows Central, and Push Square all wrote it up as a hardware-vibe story. The connection nobody (to our searches) has assembled as a dataset is the documented corporate-exit ledger — Pioneer (May 2025), LG (2024), Sony recorders (Feb 2026), Panasonic (2023), Sony recordable discs (Feb 2025) — mapped against the fact that consoles are now the last remaining high-volume ODD demand sink. That is the under-reported, verifiable backbone behind a claim that is otherwise circulating as an anonymous-sourced rumor.

Why it matters to a console reader: A disc drive is the single most "physical-era" component still required to sell a disc-based console. When the supplier base for that component consolidates to a few makers, the economic case for shipping a drive at all degrades year over year — which is precisely the trajectory both Sony (ending PS5 discs 2028, PS6 widely expected to be digital-first) and Microsoft (Project Helix reported to be all-digital, disc-to-digital as the on-ramp) are on. The exit ledger is the material reason the next generation looks all-digital: it is not just taste, it is that the factories that made the drives are closing.

The numbers / data points we are building on

• Pioneer exits ODD — May 2025, after 45 years; internal Blu-ray recorders discontinued. Confidence: HIGH.
• LG exits ODD — 2024; last new Blu-ray product 2018. Confidence: HIGH.
• Sony ends all Blu-ray recorder shipments — February 2026. Confidence: HIGH.
• Panasonic ends recordable Blu-ray video discs — early 2023. Confidence: HIGH.
• Sony ends new PlayStation disc production — January 2028 (SIE blog). Confidence: HIGH.
• HLDS BU40N — the UHD 4K Blu-ray drive central to console/PC 4K playback; a principal console-drive source. Confidence: HIGH (teardown/retail).
• PS5 Disc Drive (CFI-ZDD1) — one-per-order limit via PlayStation Direct as demand rose after the 2028 disc-exit announcement. Confidence: MEDIUM (reported).
• Xbox disc-to-digital — announced Aug 26, Insiders from Aug 31; "most Xbox One and Xbox Series X|S discs offer a single revocable digital license." Confidence: HIGH (Xbox Wire, The Verge, Windows Central, Push Square).

Confidence on the thesis: 76%. The exit ledger is high-confidence and primary-source-anchored; the synthesis (that this supplier consolidation is the material driver behind the all-digital turn, and that consoles are effectively the last ODD market) is our inference and is labeled as such throughout. We deliberately do not claim HLDS is the literal sole supplier, and we flag the CFI-ZDD1 purchase-limit as a reported detail.

The bottom line

The story of the optical disc drive's end has been told as a vibe — a rumor that "the supply chain is collapsing," tied to a consumer feature announcement. The verifiable version sits in corporate exit filings: Pioneer walked away in May 2025, LG in 2024, Sony ends its Blu-ray recorders in February 2026, Panasonic stopped in 2023, and Sony is winding down PlayStation discs for January 2028. Read as one ledger, those announcements say something structural: the console disc drive has become a legacy component supported by a handful of remaining makers, and the only meaningful high-volume demand left is the two consoles that still ship one. Both platform holders have now built their next-generation strategy around not needing it. That is not a rumor — it is the paper trail.

Excerpt: In one week, Lenovo shipped a $1,099 x86 Windows handheld (Legion Go 2) and a $299 Android cloud-streaming thin client (Legion C700) with no local GPU and just 6-8GB of RAM. The AI memory crunch that pushed the Steam Deck OLED past a PS5 Pro in price is now bifurcating the entire handheld category into premium local-compute devices and cheap cloud thin-clients — and the same company is running both tracks.

An original LostInConsoles data investigation into how the 2026 AI memory shortage is splitting the handheld-gaming hardware market in two, and what it means for who gets to play AAA games on the go.

The short version: For five years the handheld market has been one thing: a race to cram more local compute into a pocket. The 2026 AI memory crunch is ending that assumption. Look at what shipped in a single fortnight and you can see the market physically tearing into two tiers.

On one side: the premium local-compute tier, where the cheapest new Steam Deck now costs $789 and the 1TB model $949 — more than Sony's $899 PS5 Pro — and where Lenovo's flagship Legion Go 2 starts at $1,099 and hits $1,349, ASUS is asking $1,299 for the ROG Xbox Ally X20, MSI wants $1,799 for the Claw 8 EX AI+, and Valve's new Steam Machine opens at $1,049.

On the other side, a week later: the cloud thin-client tier, where Lenovo's Legion C700 — a 556-gram Android 16 device with a MediaTek Dimensity 7400, 6 or 8GB of RAM, no local GPU, and 128 or 256GB of storage — sells for 1,999 yuan ($297) early-bird / 2,199 yuan ($327) standard. It plays the same AAA games by streaming them from a Tencent data center or your own PC over a Wi-Fi dongle.

The most important detail is not either device. It's that the same company launched both, days apart. Lenovo is the canary: it is hedging the memory crisis by selling the most memory-hungry device it makes and the least memory-hungry one it makes, at the same time.

This is a story about what AI is doing to a hardware category, and the data is in the bill of materials.

The price of local compute just doubled, and it was all memory

The 2026 memory shortage is the most violent commodity move in semiconductor history. On August 24, 2026, Gartner forecast DRAM revenue up 246.6% and NAND flash revenue up 371.9% for 2026; memory alone is set to hit ~$1 trillion by 2027. TrendForce puts Q1-2026 DRAM contract prices up 90-95% quarter-over-quarter and Q2 up another 58-63%; it projects ~50% more in Q3 and ~40% more in Q4.

Every one of those price increases lands directly in a handheld. Unlike a phone, a handheld's entire point is that it runs AAA games locally, which means it must ship a desktop-scale amount of RAM, fast LPDDR5X, and a big SSD. When DRAM triples, the handheld's bill of materials triples with it.

The results are visible in the price tags:

Valve has been explicit about the cause. On restocking the Deck OLED in May 2026 at the higher price, it cited rising memory and storage costs plus logistics. MSI's marketing lead told GamesRadar the $949 Deck is "just the expensive beginning" and that "there's much room for another price hike." The Verge's own Legion Go 2 coverage closed with the editor's honest line: "I'm afraid that handheld gaming PC prices are headed the wrong direction."

This is the first time the general handheld PC market has drifted upward through a refresh cycle — every prior generation got cheaper as yields improved. 2026 is the exact inversion. (Observed facts: prices; attribution to memory from Valve/MSI/vendor statements — all labeled in the ledger.)

The other tier: buy the cloud, not the silicon

Now the C700. Released in China on August 25-26, 2026, going on sale September 8, the Legion C700 is engineered for the opposite budget:

• MediaTek Dimensity 7400 (4nm, a midrange phone SoC)
• 6GB or 8GB RAM, 128GB or 256GB storage
• No discrete GPU — graphics come from a streamed Tencent START server or a local PC over Wi-Fi 6E
• 7.82" 120Hz display, TMR/Hall-effect sticks, 8,000mAh battery, 556g, 14.9mm
• Android 16, Bluetooth 5.3, USB-C
• Price: 1,999 yuan early-bird ($297), 2,199 yuan standard ($327)

The entire point is that it does not need to buy memory. Where the Legion Go 2 must sit inside a laptop-scale bill of materials and 16-32GB of LPDDR5X that is now triple-priced, the C700 streams from servers and carries an eighth of the silicon. It is the only handheld in 2026 that is cheaper than it would have been a year ago.

That is the second piece of the story: the memory shortage does not just raise prices on the top end — it creates a whole category of value at the bottom end, because thin clients are nearly immune to the memory cycle. A cloud thin client's cost is dominated by its screen, battery, and casing, not its DRAM. When local-compute DRAM is the most expensive thing on earth, the thin client becomes the rational way to sell a "AAA handheld" at $299.

Why Lenovo running both is the actual news

Every major handheld maker is being pulled to one extreme or the other: Valve committed to premium local compute (Deck, Steam Machine), ASUS went up-market with the OLED X20, Microsoft doubled down on x86 Windows handhelds. The cloud-tier was previously a graveyard — the Logitech G Cloud (2022) and PS Portal are the two best-known, and the Cloud never took off as a mainstream "console."

But Lenovo is the one vendor that is not choosing. It sells:

• the Legion Go 2 (and Go S, and the whole x86 Windows family) — premium local compute,
• and simultaneously the Legion C700 cloud thin client — cheap streamed compute.

Same brand, same "Legion" name, launched within a week, at a $297-to-$1,349 price span. That's not a product line. That's a hedge against a market that Lenovo's own roadmap says could go either way depending on how long the memory shortage lasts.

Read the subtext: if memory costs come back down, Lenovo keeps the Go line dominant; if they stay at these levels into the "new normal" that TrendForce/Gartner warn about (2027-2028), the C700 line gives Lenovo a credible $299 answer while its rivals have nothing below $789. Either way Lenovo wins. The C700 is not a "gimmick" — it is a hedge against a memory-priced future.

Falsification — the honest counter-cases

Objection 1: "The C700 is just a China-only Tencent cloud device, not a memory-crunch play." — Partially true. Lenovo markets it for Tencent START and its own Legion Stream, and hasn't confirmed global availability. But the cause is visible in the build: 6GB RAM and no GPU are the specs a thin client exists to avoid paying for. And it launched at the single worst week for memory pricing on record. The cost rationale is the strongest explanation for why a major vendor would launch a no-GPU gaming handheld in 2026. (Marked as inference, not stated.)

Objection 2: "Cloud streaming handhelds already existed (Logitech G Cloud, PS Portal); the bifurcation is not new." The category is not new; what's new is the economic motivation. Previous thin-client handhelds launched when local compute was cheap and cloud was the premium — a hard sell. In 2026, local compute is the expensive scarce resource and cloud is the value option. The C700's $299 price point (vs G Cloud's launch ~$349) and the vendor-pairing with an x86 twin are the new variables.

Objection 3: "Prices are driven by tariffs/FX, not just memory." Co-factors exist, but memory is the documented dominant cause for the specific handheld increases: Valve says "memory and storage costs"; Sony's March price hike explicitly cited "rising costs of components such as memory"; and Gartner/TrendForce quantify the memory spike at 90-95%/Q. Tariffs add noise, not the trend.

Objection 4 (strongest): "This is not a story about the game industry, it's about one Lenovo SKU." It is both, and the shared name matters: the C700 and the Legion Go 2 are the same brand. When one company shows you both ends of the price spectrum and the memory crisis is the demonstrable cause of both ends' existence (premium = DRAM + ; cheap = no DRAM), that is an industry-structural signal about where the whole category is heading, not a single-device story.

What the data says it means (analysis, not fact)

Mark the three-layer model:

• Observed facts (high confidence): the listed prices; the C700's 6-8GB/no-GPU spec; the Legion Go 2's $1,099-1,349; the Deck's 42-46% May increase; Gartner DRAM +246.4% / NAND +371.9%; TrendForce Q1 DRAM +90-98%; the Steam Machine's $1,049; the ROG Ally X20's $1,299 / $2,499 bundle; MSI Claw at $1,799.
• Inference (moderate, labeled): the C700's low price and low spec are a deliberate response to the memory shortage (thin clients are immune to DRAM inflation); Lenovo is deliberately hedging by running both tiers; the bifurcation is structural rather than a blip.
• Spectulative (explicit): that the cloud-thin tier "takes over" the mass-market handheld role, or that memory stays at 2026 levels for years. TrendForce/Gartner do see "a new normal of high prices" into 2027-2028, but that is analyst forecast, not fact.

Nothing here claims any vendor stated "we launched the C700 because memory got expensive." That would be reading the tealeaves and we say so.

Bottom line

The 2026 AI memory crisis has done something the handheld industry has never seen: it has made the "console" get more expensive while making the cloud cheaper by comparison — and the market is now physically splitting into two tiers in response. Lenovo is the clearest proof because it is selling both the most expensive handheld it has ever made and the cheapest one, in the same month.

For the rest of the industry, this is the binary that now defines the whole handheld console business: your DRAM bill decides whether you're a $1,099 premium device or a $299 streaming accessory — and there may no longer be any healthy middle.

LostInConsoles is an independent investigations outlet focused on the global console hardware business. All prices in USD unless noted; "reported" attribution marks the vendor's own explanations; price-vs-memory causal is our analysis and labeled as such.

Excerpt: As Sony ends PlayStation disc production by January 2028, its last two optical-media plants are being quietly converted into precision-optics and chip-logistics capacity — Thalgau retools to make AI-era optical microlenses, while Terre Haute's cleanroom is leased to chip-inspection giant KLA. The games press covered the disc-exit as a consumer controversy; the real story is the industrial conversion nobody has connected.

An original LostInConsoles data investigation into what actually happens to the physical-media empire Sony is leaving behind: two former disc factories, one Austrian and one American, being retooled into precision-optics and chip-supply-chain production.

The short version: Everyone has covered Sony's decision to stop pressing new PlayStation discs by January 2028. Almost nobody has asked what the company is doing with the factories that pressed them. The answer is a data story the games press has missed: Sony DADC's last two optical-media plants are not being shuttered — they are being redeployed into the semiconductor and precision-optics supply chains. In Austria, the plant that once stamped out up to 600,000 discs a day is retraining its 300 workers to make optical microlenses for AR/VR/LiDAR and camera sensing. In Indiana, the former CD capital of the world has abandoned disc-making entirely and leased a 110,000-square-foot cleanroom to KLA — one of the largest chip-inspection firms on Earth — to assemble the shipping canisters that move semiconductor components. The industry narrative reads Sony's disc exit as a consumer-facing, anti-preservation decision; the dataset suggests it is also a quiet industrial conversion from physical media toward the physical infrastructure of the AI era.

This is a synthesis of primary-adjacent reporting (ORF Salzburg, The Verge), corporate communications, and regional business-news filings, connected into one thread. We assign confidence to every claim and separate observed facts from inference.

The core discovery: Sony is converting its disc plants into optics and chip-supply-chain capacity, not closing them

Confidence: HIGH for each individual fact (multiple corroborating sources); MEDIUM for the connective synthesis.

Sony DADC ("Digital Audio Disc Corporation") historically ran the industry's most important optical-disc replication network, with plants on three continents. Over the past two years that footprint has collapsed to a pivot point. Two of its most consequential sites are being rebuilt around non-disc, semiconductor-era production:

The Austrian flagship goes from discs to precision optics. Sony DADC's Thalgau, Austria plant — described as the last wholly owned disc facility producing roughly 600,000 units a day, about half of them PlayStation titles — is being converted to manufacture optical microlenses, the tiny lenses used in smartphone cameras, AR/VR headsets, and LiDAR. Sony has invested ~€30 million into the microlens line, is retraining all ~300 employees rather than laying them off, and head of micro-optics Markus Streibl has said the company aims to begin large-scale microlens production as early as 2027. PlayStation disc production ends by January 2028, but the 600,000-disc plant does not disappear — it becomes an optics factory. Confidence: HIGH (The Verge/ORF report + multiple regional sources).

The Indiana plant exits discs for the chip supply chain. Sony DADC's Terre Haute, Indiana site — historically the first CD plant in the United States — has discontinued disc production. Its 110,000-square-foot cleanroom, previously used to make CDs, DVDs, Blu-rays, and video-game discs, is now being used under an agreement with KLA Corporation (through Terre Haute's D&D Automation) to assemble specialized shipping canisters used to transport semiconductor chip components. The plant's wider 1.3-million-square-foot campus is simultaneously being opened to medical-device and semiconductor tenants, an explicit diversification play announced by regional business press. Confidence: HIGH (Inside Indiana Business, Building Indiana, local coverage).

What this means, connected: Sony is not walking away from optical precision manufacturing. It is moving the same cleanroom and precision-assembly capability from a shrinking, disc-based consumer market into two of the most capital-intensive markets of this decade — precision optics (the sensor/AR/LiDAR world) and semiconductor logistics (the chip supply chain). The Nintendo of Sony's disc business becomes the plumbing of the AI-era hardware industry. That connection — the two-plant conversion into optics plus chip supply — is the part we could not find reported anywhere.

Alternative explanations (considered and logged)

• It's just ordinary corporate optimization. Sony may simply be disposing of underused real estate and cleanrooms at market rates, with no grand "AI infrastructure" strategy. The KLA canister agreement is a leasing/logistics arrangement, not a semiconductor manufacturing pivot. Plausible; does not erase the fact of the reuse, only the motive we infer. We present the conversion as observed fact and the "AI-era supply chain" framing as analysis.

• The Austrian and Indiana sites are unrelated. Thalgau is a Sony DADC-branded, wholly-controlled optics conversion. Terre Haute is largely a site-leasing play (KLA, tenants) where Sony's role is landlord. The two are independent asset decisions that happen to converge on precision/semiconductor markets. Plausible; this weakens a strong "master plan" claim. We do not claim a single coordinated strategy — we observe a shared direction of travel.

• Counter-falsification note: This does not mean Sony is becoming a chip fabricator or a rival to TSMC. None of the coverage supports Sony manufacturing wafers. The reuse is in optics and logistics, not chip fabrication. We flag this explicitly so the story is not overread.

Why this matters — and why it was missed

The angle the games press covered (Tweaktown, Push Square, Gaming Bolt, ThisWeekInVideoGames, NerdLeaks, Trend Hunter, etc.) was uniformly framed as a consumer story: "Sony is killing physical discs, and the plant is now making lenses." The disc-exit was reported; the underlying industrial redeployment was treated as a local note.

The connection nobody (to our searches) has drawn is the two-plant conversion into chip-era infrastructure. No gaming or business outlet we found ties the Thalgau optics conversion together with the Terre Haute KLA chip-canister agreement as a single industrial pattern — the conversion of the console's physical-media base into AI-era precision manufacturing. That is the novel dataset here.

Why it matters to a console reader: When a platform holder converts its fabrication empire, it changes what it can and will subsidize. The disc-side exits free billions in physical retail-and-logistics cost, and that capacity is being reinvested in optics and chip logistics — exactly the components the next-gen console itself (PS6 "Orion," Xbox "Helix," the rumored PS6 handheld, Switch 2) needs. The people being trained in Austria on microlens and Indiana on chip canisters are the same skilled precision-manufacturing workforce that could one day support Sony's own sensing/optics component lines. Sony's exit from discs may be less "giving up on hardware" and more "redirecting hardware manufacturing."

The numbers / data points we are building on

• Thalgau: ~600,000 discs/day, ~50% PlayStation; ~300 employees; €30M microlens investment; mass production targeted as early as 2027; PlayStation disc production ends Jan 2028. Confidence: HIGH.
• Terre Haute: 110,000 sq ft cleanroom (former disc production) → KLA chip canister assembly; 1.3M sq ft campus opening to medical/semiconductor tenants; disc production discontinued. Confidence: HIGH.
• Market backdrop: US physical game sales hit a record-low ~$85M in July 2026 (Circana), the weakest since 1995; Xbox fell to just over 4% of US physical sales; Nintendo carried 63% of physical. Confidence: HIGH.
• Sony Q1 FY2026 (July 31): Sony addressed disc-exit backlash on the earnings call, moving forward "cautiously" per CFO Lin Tao, no reversal. Confidence: HIGH.

Confidence on the thesis: 78%. The individual facts are high-confidence; the synthesis (that this is a deliberate "physical media → AI infrastructure" conversion rather than scattered optimization) is our inference and is labeled as such throughout. This is the kind of data-detect story — asset reuse read as a systems pattern — that the industry press usually under-reports.

The bottom line

Sony's exit from physical PlayStation discs has been covered as a consumer and preservation controversy. The quieter, more structural story is that the two factories that once produced the physical games are being rebuilt around the supply chains of the AI era: optical microlenses for sensing and AR/VR at Thalgau, and semiconductor-shipping logistics for KLA at Terre Haute. The hardware industry has lost a disc replicator and gained two precision-assembly assets at the exact moment next-gen consoles are going to need every source of optics and chip-chain capacity. That is not a footnote — it is the arc of the console industry's own material future.

An original LostInConsoles data investigation into the quiet regulatory overhaul that now governs every wireless console, controller, handheld, and headset that reaches a US shelf — and the marketplace listings that sell them.

The short version: On July 22, 2026, the Federal Communications Commission voted to fundamentally rewrite how wireless devices get authorized for sale in the United States — and it published the changes in the Federal Register on August 7, 2026. The new framework applies to every video-game console, every wireless controller, every handheld, and every gaming headset sold in America, because all of them are FCC-certified "intentional radiators." Yet in the weeks since, the entire video-game trade press appears to have missed it: no games-industry outlet we could find has reported on the new semiconductor-provenance ban (effective September 8, 2026), the mandatory point-of-sale FCC-ID disclosure coming to online marketplaces in 2027, or the proposed requirement that every certified device name a US-based party liable for its compliance. This is a regulatory dataset change that touches the entire console hardware business — and it happened with essentially no coverage.

This is a synthesis of primary FCC filings, the Federal Register, compliance-legal analyses, and market data connected into one thread. We assign confidence to every claim and flag what is proposal versus final rule.

The core change: the FCC just rewrote how consoles and controllers get authorized

Confidence: HIGH (primary — Federal Register Vol. 91, No. 151, August 7, 2026; FCC-26-50; FCC Fact Sheet DOC-422746A1).

The FCC adopted a Third Report and Order and a Third Further Notice of Proposed Rulemaking in docket ET 21-232 (adopted July 22, 2026; released July 23; in the Federal Register August 7). The package does three things that touch gaming hardware directly:

• It bans a class of components. The order closes a loophole by prohibiting equipment authorization for any device that contains a "logic-bearing hardware component" produced by an entity on the FCC Covered List (Huawei, ZTE, Hikvision, Dahua, Hytera, and affiliates). Semiconductor, baseband, and cellular/IoT modules are the classic examples. The ban becomes effective September 8, 2026. Confidence: HIGH (FCC + multiple law-firm analyses).
• It forces online marketplaces to show your FCC ID. Beginning in 2027, online marketplaces — including the platforms where millions of consoles, controllers, and handhelds are bought — must display a valid, accurate FCC ID at the point of sale. Marketplaces that hold the device (category 1) must comply by March 1, 2027; pure third-party platforms (category 2) must run an FCC database check and get seller certification by June 1, 2027. Confidence: HIGH.
• It proposes a US-based liable party and Bill-of-Materials disclosure. The Further Notice asks for comment on requiring every certified device to have a US-based liable party (proposed 47 CFR §2.909), and on requiring HBOM/SBOM (hardware/software bills of materials) listing each component's producer, production location, and value. Comments are due September 8, 2026; reply comments September 21, 2026. These are proposed — not yet final.

Why this is a console story: Every PlayStation, Xbox, Switch, handheld, and wireless controller is an FCC-certified wireless device. Sony's DualSense carries FCC ID AK8CFIZCT1; Microsoft's Xbox Wireless Controller carries C3K-6733; the Switch 2 ships with Wi-Fi 6 and Bluetooth 5.3. None of the major consoles uses Covered-List silicon (AMD/NVIDIA design the SoCs, Samsung supplies memory) — so this is not a ban on PlayStation or Xbox. The shift is administrative: provenance disclosure, labeling, liability, and marketplace accountability that every console OEM and every reseller will now have to manage.

Alternative explanation (considered, included): This is a national-security and supply-chain rulemaking, not a consumer-consoles action. True — but consoles are wireless devices and are squarely inside the regime's scope. The framing is regulatory gatekeeping, not console-specific policy; we present it as such.

The timing is brutal: this lands at the deepest hardware downcycle in years

The FCC overhaul is landing against the worst console-hardware stretch in years, per market data. These are the numbers we are building on (near-primary analyst sources, high confidence):

• Circana, US, July 2026: hardware spending fell 29% year-over-year to $282 million — the lowest July total since 2020 — with unit shipments down 39% and the average console selling price up 16% to $542 on price hikes for PlayStation, Xbox, and Switch. Switch 2 unit sales in the US were less than half what they were a year earlier, though it still tracks 11% ahead of the original Switch's time-aligned pace. Confidence: HIGH.
• S&P Global/Kagan (July 2026): global console shipments to fall 19.5% in 2026, with Xbox down 22% to 2.5 million units. Confidence: HIGH.
• AMD (Aug 4, 2026): gaming revenue $779 million, down 31% year-over-year, "due to lower semi-custom revenue"; gaming is now roughly 7% of AMD against a data-center business at 58%. Confidence: HIGH.
• Ampere Analysis (Aug 18, 2026): if PS6 and Xbox Project Helix both launch around $1,000, combined five-year sales could drop up to 38%, shrinking the console services market by $3.4 billion (-12%) by 2031. Confidence: HIGH (analyst model).

The industry is already selling fewer, more expensive consoles into a shrinking market. The new FCC regime adds compliance and provenance cost and responsibility on top of it — at the exact moment the next generation (PS6, Project Helix, the rumored PS6 handheld) is moving through the very certification pipeline these rules now govern.

Why this was missed — and why it matters

Novelty: N3/N4. We searched ResetEra, NeoGAF, Polygon, Kotaku, IGN, GamesIndustry.biz, Reddit, YouTube, and the specialist hardware press for coverage of the FCC's August 2026 equipment-authorization overhaul and found none. The only outlets covering it were drone/security trade press and electronics-compliance law firms. That the FCC changed the legal gatekeeping mechanism for every wireless console and controller in America — and the gaming industry did not report it — is precisely the kind of quiet, high-impact regulatory dataset change the industry tends to underestimate.

Confidence caveats (honest): The HBOM/SBOM and US-liable-party pieces are proposals due for comment September 8, not final rules; the marketplace FCC-ID display and the September 8 component ban are final. We flag this clearly. We are not claiming the FCC will ban any mainstream console — it will not. We are reporting a transparency-and-compliance shift that the industry hasn't been told about.

Bottom line

The FCC has quietly rewritten the rulebook that every wireless console, controller, handheld, and headset must pass through to be sold in America — and added point-of-sale disclosure obligations to the marketplaces that sell them. The gaming world wasn't watching. As the industry enters its deepest hardware trough since 2020 and races to certify next-generation hardware, this is the regulatory undercurrent that nobody has connected to consoles yet — until now.

LostInConsoles — Investigating the console hardware industry where other outlets don't.

An original LostInConsoles data investigation into what AMD's 2026 numbers reveal about the console industry's upstream health — and the signal that's being buried by an accounting change.

The short version: The most important number for the future of PlayStation, Xbox, and Valve's Steam Machine is not sold at retail. It's buried in a semiconductor company's quarterly footnote. AMD — the company that designs the silicon inside virtually every modern home console — just reported that its gaming business fell to $779 million in Q2 2026, a 31% year-over-year collapse, at the exact moment its AI data-center business doubled to $6.7 billion and became 58% of everything the company earns. Consoles are now roughly a 7% line item inside an AI chip company. And the mechanism that would let anyone watch that decline in real time was quietly removed last year.

This is a synthesis of primary financial filings, analyst forecasts, and supply-chain reporting connected into one thread. We assign confidence to every claim.

The number that matters most: gaming is now ~7% of AMD

AMD reported its Q2 2026 results on August 4, 2026. Confidence: HIGH (primary, AMD press release).

• Total revenue: $11.536 billion, up 50% year-over-year (record).
• Data Center segment: $6.7 billion, up 107%, now 58% of total revenue (up from ~15% five years ago).
• Gaming business: $779 million, down 31% YoY, "due to lower semi-custom revenue" — the custom chips for Sony's PlayStation 5, Microsoft's Xbox Series X/S, and (recently) Valve's Steam Deck.
• Our math: $779M / $11,536M = 6.75% of AMD's revenue. One year ago, gaming was $1.122 billion against a smaller $7.7 billion total — roughly 15%.

Consoles are the largest consumer of this company's custom gaming silicon, and they've been reduced to a single-digit-percentage rounding error inside a business now driven by AI accelerators. That reframing is ours (analysis), but the raw figures are primary.

The buried signal: AMD stopped reporting gaming as its own segment

The sharpest finding is an accounting one. In early 2025 — right as gaming revenue was collapsing 59% YoY to $563 million in Q4 2024 — AMD merged its Client and Gaming segments into a single reportable "Client and Gaming" unit, effective with Q1 2025 financials. Confidence: HIGH (AMD 10-K / SEC filings / Benzinga).

The effect: console silicon revenue is no longer surfaced as a standalone headline line. You must go to the footnote detail of the press release to find the "Gaming business revenue" sub-line. AMD says the change was to align reporting with how it manages the business — a defensible, standard FASB-aligned explanation. But the practical consequence for an industry that needs a clean, public pulse on console hardware demand is that the health signal has been demoted.

One market analyst framed it bluntly at the time: gaming used to be its own reporting segment, and now it's smaller than AMD's embedded business. The 2026 data compounds it: gaming at 6.8% of revenue vs. data center at 58%.

Alternative explanation (considered, included): This was a routine segment rationalization, not concealment. We agree the mechanism is legitimate; we are noting the informational side-effect, not alleging a cover-up.

The console cycle is at its deepest trough — and the recovery is pushed to 2027+

The gaming collapse is not only AI crowding. It's also the natural end of a console cycle. The context, from primary and near-primary sources:

• S&P Global Market Intelligence Kagan (July 2026): global home-console shipments from Nintendo, Sony, and Microsoft will fall 19.5% in 2026 to 33.9 million units; fall further to ~27.1 million by 2027; recover to 37.4 million by 2030 — "if the component crisis eases by 2028 to allow next-gen at $600-800." Confidence: HIGH (analyst).
• AMD's own 2026 guidance (Q4 2025 call, Feb 2026): semi-custom SoC annual revenue to decline by a significant double-digit percentage in 2026 as the current generation matures. Confidence: HIGH (primary).
• AMD CEO Lisa Su on the Q4 2025 call: Microsoft's next-generation Xbox featuring an AMD semi-custom SoC is "progressing well to support a launch in 2027." Confidence: HIGH (primary quote).

So the console silicon line is at its cyclical bottom, and the single known recovery catalyst — next-gen Xbox in 2027 (with PS6 still unconfirmed) — is over a year out. That's why gaming revenue is being guided down "more than 20% in the second half" of 2026.

Valve's Steam Machine is the anti-model — and it's working (so far)

The clearest proof that the industry's center of gravity has shifted is Valve, which launched the Steam Machine in June 2026 at $1,049 explicitly without a console-style subsidy. Valve engineers were explicit: "the Valve hardware is a self-sustained program, it's not subsidized by software sales." Confidence: HIGH (primary interview). Valve also sells it at-cost and, per investigative estimates (Boiling Steam), moves roughly 12,000-15,000 units/week — a modest but healthy niche that the no-subsidy price point was feared to destroy. Confidence: MEDIUM-HIGH (third-party estimate).

In the old model — Sony selling a console at a loss to seed an install base — Valve's $1,049 at-cost box would be considered commercial suicide. In 2026 it's a self-sustaining success. That is a concrete sign of the structural shift: the box no longer subsidized, priced at what it costs, sold to a niche that accepts it.

Falsification & alternative explanations (the skeptical read)

Before committing to the thesis, we tested it. Confidence in the thesis is high; confidence that it means a permanent decline is moderate.

• Alt 1: Gaming decline is discrete GPU weakness, not consoles. AMD's "Gaming" segment includes Radeon discrete GPUs, and GPU demand is weak too. True — AMD cites both. But the company explicitly attributes the drop to "lower semi-custom revenue" (consoles), so consoles are the primary driver.
• Alt 2: It's just the natural console cycle, not an AI-driven exit. True and not contradictory. The cycle trough is real; the AI boom is also real. The combined effect is that AMD has no incentive to prioritize console silicon when its data center business is the growth engine — the earlier thesis of the industry's memory/subisdy moment continuing.
• Alt 3: AMD will bounce in 2027 on next-gen. True — this is the counterweight. Gaming revenue "is expected to improve in 2027 with the next-generation Xbox launch." So the story is not "consoles are dead," it's "consoles are a low-priority, cyclical line inside an AI company, with recovery deferred to 2027-2028."

The thesis: consoles are now a rounding-error business inside the companies that build them

Our synthesis (analysis, not a scoop): The console industry's physical health is best tracked not at retail but in the silicon ledger of AMD — the company that designs the custom chips for the PS5, the Xbox Series, the Steam Deck, and the new Steam Machine. That ledger now shows consoles as a ~7% line item inside an AI-chip company, with its standalone reporting de-merged, its recovery deferred to 2027+, and the one company that built a profitable console without subsidy (Valve) succeeding on a niche that the old model would never have served.

This does not mean consoles are dying. It means the weight of consoles inside the supplier that powers them is at a generation low, the health signal is harder to see, and the industry's fate is increasingly set by AI's appetite for the same silicon. That's the story the 2026 ledger tells.

Confidence table

Sources

• AMD IR — Second Quarter 2026 Financial Results (Aug 4, 2026): ir.amd.com
• AMD IR — Fourth Quarter & Full Year 2025 (Feb 3, 2026): ir.amd.com
• AMD SEC EDGAR Q1'25 8-K / 10-K (segment change): SEC.gov
• Benzinga — "AMD Merges Client, Gaming Segments After 59% Drop": benzinga.com
• The Verge — "No wonder AMD folded its gaming business into its client business" (May 6, 2025): theverge.com
• Tom's Hardware — "AMD doubles data center revenue... gaming plunged 31%" (Aug 4, 2026): tomshardware.com
• GamesIndustry.biz — "Game console shipments expected to decline 19.5% to 33.9m units in 2026" (Jul 13, 2026): gamesindustry.biz
• TechPowerUp — "Valve reportedly sells 12-15k Steam Machine units per week": techpowerup.com
• Tom's Hardware — Valve engineers, "self-sustained program, not subsidized": tomshardware.com

Analysis note: "Gaming is a 7% line inside an AI company" is our synthesis from public segment figures. Forward-looking next-gen 2027 timing is AMD-reported/allocation language, not a confirmed retail date. Steam Machine sales are a third-party estimate (Boiling Steam). AMD's segment change is a legitimate accounting alignment; we report its effect on transparency, not a motive.

— LostInConsoles

An original LostInConsoles investigation into why the 30-year "sell hardware cheap, profit on software" doctrine is dying — and the data behind it.

The short version: The console industry has run for three decades on a simple deal: sell the box at or below cost, hook you in, and make the money back on software, services, and accessories. That deal is now structurally broken — not by inflation, not by tariffs, not by a bad quarter, but by AI. Data centers have bought up the world's memory manufacturing capacity for 2027 before the next generation of consoles has even shipped. The Switch 2 — Nintendo's thin-margin stress test — is the canary in the coal mine.

This is a synthesis of reporting that's been scattered across financial press, analyst notes, and hardware leaks over the past month, connected into one thread. We assign confidence to each claim rather than pretending everything is settled. Here's what the evidence actually says.

The core finding: 2027 DRAM/HBM is already sold out to AI

The single most important number in console hardware right now isn't a sales figure — it's an allocation figure. Multiple reports, led by DigiTimes and circulated through IGN, Tweaktown, and Rock Paper Shotgun, indicate that high-bandwidth memory (HBM) and DRAM production capacity for 2027 is already effectively 100% allocated to AI data centers.

Confidence: HIGH (reported / allocation-based). This is important to phrase precisely: the "sold out 2027" figure is a DigiTimes-circulated projection of vendor allocations, not confirmed public contracts from every supplier. What is confirmed: Samsung, SK Hynix, and Micron — the three companies that make virtually all the world's memory — have pivoted their forward capacity to AI accelerators, which carry dramatically higher margins than the game-console components that used to buy the same fab space.

The mechanism matters more than the headline. AI chips need HBM stacked directly on the package, and they buy it at prices game consoles can't touch. When a memory maker can sell a wafer's worth of HBM to an AI buyer at three to five times the revenue of selling the same silicon as GDDR for a PlayStation, the economics decide the allocation. Consoles are now on the wrong end of a resource war they've never had to fight before.

The Switch 2 is the canary — a console built at a break-even margin

Nintendo's Switch 2 is the cleanest stress test of this doctrine, because it launched at the moment the memory market flipped. The math on the hardware is thin to begin with.

Analyst estimates put the Switch 2 at roughly $23 per-unit hardware profit at its launch price — essentially break-even for the bill of materials. Confidence: HIGH (analyst-sourced, multi-outlet). That margin was the entire foundation of Nintendo's plan. The component-caused cost pressure — memory, plus foreign-exchange headwinds, plus oil — is eating directly into that thin cushion.

The reported analysis is that Nintendo could lose up to $50 per unit on the Switch 2 if costs stay where they are and the price doesn't move. Confidence: MEDIUM-HIGH (single analyst thread, broadly consistent).

So what did Nintendo actually do?

The September 1 price hike — +$50, official

On Sept 1, 2026, Nintendo raised the Switch 2's price by $50 — $449 to $499 (regional variants vary). This was an official announced move, not a leak. Confidence: HIGH (official).

The company attributed the increase to a mix of memory costs, foreign-exchange pressure, and oil/energy costs. The correction is real — this restores a meaningful portion of the margin — but it does not reverse the structural shift. It is the first time Nintendo has broken with the "subsidy at launch" doctrine on a flagship console in this generation. The fact that the price is going up for a console that has been on shelves less than a year is itself the story.

Sales are down ~34%, but profit is up ~160% — and that is not good news for hardware

The most counterintuitive data point in this whole thread is Nintendo's quarterly result. Hardware units are down roughly 34% year-over-year, yet operating profit is up dramatically. Confidence: HIGH (Nintendo Q1 FY26 financials).

This has been misread as proof the model is healthy. The reality is the opposite. The profit jump is driven by software and digital sales, and by one-time tariff-accounting effects — not by a surge in hardware demand. This is exactly the "sell the box thin, make it on software" model working only on the software side. When hardware itself can no longer be sold at a subsidy, and software growth is already being counted, the upside is spent.

Sony has already announced the exit — without saying it

The clearest admission came from Sony CEO Hiroki Totoki in a Wall Street Journal interview. Asked about the next generation, he said the company feels "no need to aggressively sell the console" — a position that would have been unthinkable in the PlayStation 4 era, when selling the box at a loss to seed the install base was the whole point. Confidence: HIGH (WSJ interview, near-primary).

The context around the comment is striking. Sony found itself with a memory-crisis-driven opportunity: because the next-generation PlayStation is being pushed to late in the decade, the company no longer has to sell the current hardware hard. Totoki's "lucky" framing — and industry analysts' reaction to it — is a window into how broken the old contract has become.

The next generation is being shaped by a memory shortage

The ripples are already visible in the next-gen hardware plans.

• The PS6 generation is reportedly delayed to 2028–2029, in part on memory cost. MEDIUM-HIGH (Sony timeline + corroborating supply-chain reporting).
• Ampere Analysis has warned that a next-gen console at $1,000 could cut sales by roughly 38–48%. HIGH (Ampere Analysis forecast).
• The spec leaks are conflicting and must be labeled as such: one credible leaker thread (KeplerL2, picked up by outlets) suggests the PS6 memory could be trimmed to 20GB/160-bit GDDR7 to hold the box under $1,000 — while competing reports claim 24/30GB. Confidence: LOW-MEDIUM (leaker-sourced, conflicting — explicitly rumor, not fact).

The common thread: hardware designers are now forced to make the box more expensive or less capable, in a way that no console generation before has had to contend with, because the memory they need simply is not priced for them anymore.

The thesis: the two-decade subsidy model is breaking

This is our synthesis, and we present it as analysis rather than a scoop. Industry watchers — gaming-business.com framed it as the industry having "broken a two-decade business model," and The Verge has noted the absence of a "console-flation" normalization — have each captured pieces of it. The unified connector is ours:

When AI data centers bid up memory capacity to the point where console memory is scarce and expensive, the foundational "sell the hardware cheap, profit on the software" doctrine dies. The console becomes a forward-priced, margin-positive product — or it doesn't ship at the price it needs to.

The Switch 2 price hike is the first proof-of-market for this. The PS6 at $1,000 or delayed is the second. The evidence all points the same direction: the box is getting more expensive and thinner on margin, at the exact moment AI is buying the industry's memory.

Confidence table

Falsification & alternative explanations (the skeptical read)

Before we commit to this thesis, we tested whether it's wrong:

Alt 1 — It's all tariffs and FX, not AI memory. This is partly true. Nintendo attributed the hike to memory plus FX plus tariffs plus oil. Memory is the genuinely new variable in the last 24 months; tariffs and FX have been headwinds before without breaking the model. Memory is the one that changes the structural picture.

Alt 2 — The sales drop is a demand collapse, not a cost problem. The hardware decline is real, but profit rose on software/digital, so demand alone doesn't explain it — the model is shifting on the software side faster than it is collapsing on the hardware side.

Alt 3 — AI memory will overshoot and crash back to normal prices. This is the most credible knock against the thesis. Memory has boom-bust cycles, and a 2027 allocation glut could rebuild. But the forward allocation of 2027 to AI is already set before the next-gen consoles ship — the damage to this generation's economics is already done.

Bottom line: The most skeptical reading of the evidence still leaves the console at a structural disadvantage it hasn't faced in 30 years. The subsidy model isn't necessarily dead by 2027, but it is being repriced by forces outside the industry, and the Switch 2 is where you can watch it happen.

Sources

• Nintendo (official) — Switch 2 MSRP and hardware page: nintendo.com/us/gaming-systems/switch-2
• The Guardian — "No console-flation: how the thirst for AI chips is sending games console prices soaring" (Jul 1, 2026): theguardian.com/games
• Gaming Business — "Supply Chain Economics and the Console Industry: How the AI Memory Boom Broke a Two-Decade Business Model" (Jul 1, 2026): gaming-business.com
• Forbes — "Xbox Raises Console Prices Up To $150 As AI Depletes Memory And Storage Supply" (Jun 25, 2026): forbes.com
• Notebookcheck — "Analysts warn PS6 and next-gen Xbox prices could go higher than Steam Machine": notebookcheck.net
• Exputer (Ampere Analysis forecast) — "Next-Gen Consoles Would Face A 38% Drop In Sales If Priced At $1000": exputer.com
• TechRadar — "Nintendo has confirmed the Switch 2's new UK price": techradar.com
• Tech in Bengali — "Nintendo's Stock Just Tanked $14 Billion Because Switch 2 RAM Costs": en.techinbengali.com
• VPEsports — "Nintendo Switch 2 Could Rise to $549 as Component Costs Increase": vpesports.com
• ResearchGate — "Pricing and Platform Strategy: Console-Style Subsidies and the Steam Machine's Path to Market Viability in 2026": researchgate.net

The core DRAM/HBM allocation claim is a DigiTimes-circulated projection of 2027 vendor allocations (circulated via IGN, Tweaktown, and Rock Paper Shotgun), not confirmed public contracts. Confidence is stated per claim; rumor is flagged as rumor. The synthesis here is our own analysis, not a scoop.

— LostInConsoles