A frontier lab is forced to divest a business

How I read the question

YES needs three things together. (1) A divestiture of a business that is mandatory and binding. (2) It must come from a court order or from a settlement, commitments or remedies accepted by a competition authority. A unilateral administrative order does not count. (3) The decision itself, or reasoning it incorporates, must expressly name a market for AI models, AI chips or AI compute. Mentioning AI as a trend is not enough. Voluntary sales, sales of minority stakes and behavioural remedies do not count. Only publications dated 16 September 2026 or later count.

Where things stand (as of 17 October 2026)

I found no qualifying publication.

Google ad tech (Eastern District of Virginia). On 2 September 2026 Judge Brinkema refused to order divestiture of AdX, DFP or anything else, and imposed behavioural remedies plus a monitor (Ashurst "The Bug", Sept 2026; opinion, ECF 1858). I downloaded the 106-page opinion and searched it. "AI" appears 25 times, always as a technology or industry trend. The phrases "AI market" and "market for AI" appear zero times. So even a later divestiture on appeal would almost certainly fail the market test.

Google search (D.C. Circuit). The Justice Department's cross-appeal brief of 28 July 2026 challenges only the rejected ban on default payments. It does not appeal the refusal to order a Chrome sale (DOJ brief). That matters because Judge Mehta's remedies opinion has a section titled "The GenAI Market" (ECF 1436). A Chrome divestiture would probably have met the market test, so this path is now essentially closed.

FTC v. Meta. The FTC lost at trial in 2025 and is appealing. No argument date has been set (MediaPost, 5 Oct 2026). It is also a social-media case.

Europe. The European Commission is still assessing Google's ad tech compliance plan, and the complainants no longer expect a breakup (The Next Web, Sept 2026). The Commission's interim measures on Meta's WhatsApp do name an "AI assistant market", but the remedy is an access obligation, which is excluded (Tech Policy Press, 6 Aug 2026).

Recent deals. None left a remedy behind:

• China's state planner ordered Meta's Manus deal unwound under foreign-investment rules, not competition law, and before the window opened (April 2026).
• SpaceX/Cursor closed without conditions on 14 August 2026.
• Google/Wiz and Alphabet/Intersect closed in March 2026.

I found no pending in-depth merger review involving any covered group.

Base rate

Across roughly 20 years and about 10 in-depth merger reviews of Alphabet, Meta, Alibaba, ByteDance and the Musk companies, there has been no qualifying divestiture. Meta/Giphy was a unilateral order by the UK regulator. Across big tech more broadly, divestiture remedies are rare. My rough rate for these groups is about 1.5–2.5% a year for any qualifying divestiture. About 40–45% of those would name an AI market; that is likely for AI-related deals and unlikely for grocery, security or telecom deals.

Paths to YES (5-year estimates)

• Conditional merger clearances, about 3%. Candidates include Alphabet, SpaceX (now an AI compute seller), Meta, OpenAI after its IPO, Anthropic, Alibaba and ByteDance. Since 2025 the US agencies have again been accepting negotiated remedies, and China's competition regulator imposed tech divestitures in 2025 (China Briefing).
• A Tesla–SpaceX merger, about 0.5%. Prediction markets put a deal at 24% by December 2026 and 63% by end-2027 (BigGo, Oct 2026). But the two companies' businesses barely overlap, so a required sale is unlikely.
• FTC v. Meta revived and ending in divestiture, about 0.4%.
• New conduct cases, EU digital-markets commitments, or other courts, about 0.4%.

Timing

The near term is very low because nothing is pending and merger reviews take 6–18 months. I set a quarterly hazard of about 0.2–0.27% from 2028 onward, rising slightly as AI deal-making grows. Compounding the 21 quarterly hazards gives about 4.25% by end-2031.

Strongest case against my estimate

Five years is long. Big AI deal-making is accelerating (SpaceX's $60 billion Cursor purchase, Alphabet's $32 billion Wiz deal), and decisions now routinely name AI markets. A single remedy-laden deal would flip the answer. I considered going to 6–8%. I stayed near 4% because of the zero track record, three failed breakup attempts in 2025–26, the dropped Chrome appeal, and how AI deals are now structured to avoid merger review.

Question. By the last day of each quarter, has a court order, or a settlement with a competition authority, required a covered developer or its parent to divest a business, in a decision that names a market for AI models, AI chips, or AI compute?

Definitions.

• Covered developer: the site convention applies. Parent means an entity directly or indirectly controlling the developer when the order or settlement is issued; this includes the corporate owner of a developer operating as a division.
• Court order: an entered judicial order or judgment in any jurisdiction, including a court-approved consent decree. Settlement with a competition authority: a binding agreement or undertaking formally accepted by a public authority empowered to enforce competition or merger law, including a consent order or merger-clearance decision incorporating agreed divestiture commitments. A unilateral administrative order without a settlement does not qualify unless a court itself orders the divestiture.
• Divest a business: a mandatory sale, spin-off, or transfer of a subsidiary, operating business unit, or package of operating assets constituting a business, such that the developer and its corporate group relinquish control of that business. The business may be held through a subsidiary and need not itself supply AI products. A sale required as a binding condition of permission to complete a specified acquisition counts; mere permission to divest as one optional remedy does not.
• Names a market: the order, accepted settlement, or accompanying decision or opinion expressly identifies a market for supplying AI models (including foundation models and model API services), chips used for AI training or inference, or computing capacity/services for AI training or inference. A formally adjudicated relevant-market definition is not required. General semiconductor or cloud markets qualify only if the decision expressly identifies an AI-specific market or submarket. Mentioning AI as a technology, customer use case, competitor, or future trend without identifying such a market does not qualify. Incorporated reasoning in an earlier opinion counts; an allegation quoted but expressly rejected does not.

Resolves YES if an official publication within the resolution window establishes both the binding divestiture requirement and the named AI market above. The named market must appear in the decision imposing or accepting the remedy or its incorporated reasoning, not merely elsewhere in the case record. The divested business need not be in that market, and no finding of monopolisation is required. Entry or formal acceptance is sufficient: completion of the sale and exhaustion of appeals are not required. A stay pending appeal does not disqualify an entered order. Later reversal or abandonment does not undo a YES, under the resolution-window convention.

Exclusions (resolve NO).

• Complaints, proposed remedies, draft decrees awaiting required approval, voluntary sales without a binding requirement, or settlements announced only as agreements in principle.
• Fines, data sharing, licensing, interoperability, access obligations, internal ring-fencing, temporary hold-separate arrangements, and sales of isolated assets or noncontrolling investments that do not transfer control of an operating business.
• Merger prohibitions that do not require divestiture, or divestiture triggered only by a future violation that has not occurred.
• Search, advertising, social-media, or other remedies that mention AI but do not name an AI market as defined above. Google search behavioral remedies and Meta/Giphy divestiture do not qualify merely because their subjects are covered corporate groups.
• No qualifying publication by the relevant quarter-end.

Resolving source. The issuing court's entered judgment, order, consent decree, or accompanying opinion, or the competition authority's published accepted settlement, undertakings, consent order, or commitment decision, available from its official website or public docket. Eligible authorities include the US Department of Justice and Federal Trade Commission, state competition enforcers, the European Commission, the UK Competition and Markets Authority, China's State Administration for Market Regulation, and equivalent competition authorities in other jurisdictions. These primary legal documents are the exclusive resolving sources, narrower than Public confirmation (A2); company statements and press summaries alone do not suffice. Apply Resolution date (A3) and the resolution-window convention. Background examples: DOJ Google search remedies, 2 September 2025; CMA Meta/Giphy case record.