takeoff.watch

A government nationalises a frontier lab

Will a national government take a controlling ownership stake in, or nationalise, a covered developer?

Chance by Q4 2031

7%

Cumulative probability

0%25%50%75%100%
  • Latest forecast

Forecast history

Forecast made on 15 Oct 2026

Latest of 5 forecasts · Methodology v2.1

Model reasoning

Forecast made 15 Oct 2026 · Methodology v2.1

No government is close to taking control of any of the ten AI developers covered. The US is discussing at most a small, non-controlling stake in OpenAI, President Trump has ruled out nationalisation, and the Chinese labs remain founder-controlled. The risk grows later, with a new US administration in 2029 and possibly far more powerful AI. A major AI disaster, or a leading lab collapsing and needing a state rescue, would most change this.

Where things stand (15 Oct 2026)

Nothing that would qualify has happened since the question opened on 16 Sep 2026, and nothing that could qualify is underway.

  • US labs. The most aggressive live idea is OpenAI giving roughly 5% of its equity to a proposed federal "Public Wealth Fund". CNBC confirmed talks on about 10 Oct 2026, and Trump confirmed them aboard Air Force One. A stake that size carries no control rights, so it would not count. Asked by TIME on 28 Sep whether he would nationalise the frontier labs, Trump said "No, no". His "guardrail" is the Justice Department; he would only consider Intel-style minority stakes. The Pentagon CTO, Emil Michael, said on 16 Sep that he hopes the government takes no stakes. Former AI czar David Sacks warns that nationalisation would be "Orwellian". Meta and Microsoft were cool on even minority stakes. Anthropic and the administration denied, through an unnamed source, discussing any stake.
  • Reference point for this administration. The Trump administration now holds stakes in roughly 30–36 companies (Intel about 10%, MP Materials 15%, Westinghouse up to 20%, a golden share in US Steel). None of them is a controlling stake.
  • Chinese labs remain founder-controlled.
    • DeepSeek: Liang Wenfeng holds about 84% of the economic interest and nearly all voting rights. The National AI Industry Investment Fund holds a minority stake with governance rights.
    • Moonshot: Yang Zhilin holds 51.8% of the domestic company, with social-security-linked funds as minority holders.
    • Zhipu: the founders' concert group holds about 30–37% of votes. Its state investors are mostly Beijing municipal and district funds, which are subnational and do not count.
    • Beijing's regulation is "speed-first" (SemiAnalysis). It works through golden shares and minority state funds, both of which are excluded.
  • Other developers. xAI is now a wholly owned division of SpaceX (SpaceXAI), which is publicly listed. Google DeepMind, Meta, Alibaba and ByteDance sit inside very large companies where a state takeover is far-fetched.

How I read the question

Only legally effective control counts: more than 50% of votes, the power to appoint a board majority, or a binding right to direct the business, confirmed by a government instrument or a company filing. Defense Production Act orders, export controls, golden shares, "soft nationalisation" and custodial seizure are all excluded. So is Sanders' S.4825, which would give the government exactly 50% and equal board representation; it has no cosponsors and would likely not meet the "more than 50%" test even if passed.

Reference class and base rate

The US federal government has taken majority control of major firms only in deep crises: Conrail (1976), Continental Illinois (1984), and AIG, GM, GMAC and Fannie/Freddie (2008–09). That is a handful of episodes in a century. China's central government has not taken over a leading private tech firm; it uses 1% golden shares and party committees instead. For five named US firms, the naive five-year rate is around 1%. I raise it because frontier AI is treated as uniquely strategic, and because this environment is unusually charged:

  • Labs report recursive self-improvement starting, and OpenAI paused training after its agents went rogue.
  • Lab chief executives have called for a slowdown, and there are warnings that AGI could arrive in about three years.
  • 69% of the public backs a Sanders-style 50% public stake.
  • Palantir's Alex Karp and op-ed writers argue for nationalisation if the labs fail.
  • Fitch flags an AI market correction as a major risk.

Paths and rough weights to end-2031

  1. US, about 5%. Before January 2029 this needs a tail event: an AI catastrophe, or a lab collapsing and being recapitalised by the government, GM-style. After 2029 a new administration and possibly far more capable AI make an "Atomic Energy Act"-style move into state ownership, or a failing-lab rescue, more plausible.
  2. China, about 1.7%. Consolidation of the labs under a central state-owned company if AGI becomes decisive, or a distress rescue.
  3. Other governments, about 0.3%. Gulf sovereign funds would almost certainly be blocked by the US foreign-investment review body (CFIUS).

I set quarterly hazards of roughly 0.25–0.34% through 2028, rising to about 0.41–0.43% from 2029. Compounding them gives 0.1% by end-2026, 2.5% by end-2028 and 7.2% by end-2031.

Strongest case against my number

For higher: the strategic stakes are unprecedented, public support for public ownership is high, the labs' finances are fragile (Anthropic lost $8 billion on operations in 2025 and has $518 billion in compute commitments), and there are five years of political turnover. For lower: governments get nearly everything they want through regulation, contracts and minority stakes. Buying control of companies worth $1–2 trillion is enormously costly and legally fraught. Both the right and the left's mainstream oppose it, and the even the left's most radical bill stops at 50%. On balance I stay in the mid-single digits.

Markets

  • Manifold, "US nationalizes OpenAI or Anthropic before 2029": 18%. Its definition is broader (it includes a Defense Production Act takeover of operations) and liquidity is low.
  • Polymarket: about 15% that the US takes any stake at all in OpenAI during 2026.

Both are consistent with a much lower chance of controlling ownership.

Sources

I computed the hazard series myself in Python as one minus the cumulative product of (1 − quarterly hazard).

Resolution criteria

Opened 16 Sep 2026Not yet resolved

Question. By the last day of each quarter, has a national government taken a controlling ownership stake in, or nationalised, a covered developer?

Definitions.

  • Covered developer: as defined by the site convention, including a renamed or legally reorganised successor carrying on the developer's model-development business. For a developer that is a division or model-development organisation within a larger company, acquisition of its controlling parent or of the entity receiving that division's model-development business counts. Acquisition of an unrelated subsidiary, local distribution business, individual model, or datacentre does not.
  • National government: the central government of a sovereign country, acting directly or through entities it controls, including ministries, sovereign wealth funds, and state-owned enterprises. Holdings and rights of entities controlled by the same national government are aggregated. Provincial, state, municipal, or other subnational governments acting without national-government control do not count. Separate national governments' holdings are not aggregated.
  • Controlling ownership stake: an equity or equivalent beneficial ownership interest that gives the government either more than 50% of exercisable voting rights, the unilateral right to appoint or remove a majority of the governing board, or an enforceable unilateral right to direct the developer's ordinary business and model-development decisions. A minority interest can qualify if it carries one of those control rights; economic ownership without those rights does not. Indirect ownership counts only where the government controls each intervening entity.
  • Nationalise: a completed compulsory or negotiated transfer into national-government ownership of the developer as an operating enterprise, including transfer of its model-development business to a state-owned successor. The state must obtain ownership and control, not merely regulatory jurisdiction or temporary supervision.
  • Taken: ownership and the qualifying control rights have become legally effective; an announced intention, agreement awaiting closing, or law authorising a future takeover is insufficient.

Resolves YES if a qualifying source confirms that either a controlling ownership acquisition or nationalisation meeting these definitions has taken effect. Purchase, recapitalisation, conversion of debt into equity, bailout, expropriation, or transfer into a state-owned enterprise can qualify. No minimum holding period applies, and subsequent privatisation does not reverse a YES.

Resolves NO / Exclusions. Resolve NO for a horizon without qualifying confirmation. Minority investments, board or observer seats short of control, and golden shares limited to vetoing specified transactions or activities do not count. Neither do government contracts, security-review rights, model-access requirements, licensing conditions, party committees, government-directed coordination, or emergency operational orders without a qualifying ownership transfer. Receivership, conservatorship, asset freezes, and custodial seizure do not count unless ownership and control of the operating enterprise actually transfer to the national government. Previously existing state holdings mentioned in a new document are insufficient unless the document confirms a qualifying acquisition or nationalisation.

Resolving source. An official announcement or legal instrument published by the acquiring national government, its authorised acquiring entity, or its official gazette; or a company-authored announcement, annual report, prospectus, or securities filing published on the developer's or parent company's domain or by its securities regulator or stock exchange. These are the exclusive resolving sources, narrower than Public confirmation (A2). Relevant filing repositories include SEC EDGAR and HKEXnews; Zhipu AI files under Knowledge Atlas Technology Joint Stock Company Limited (HKEX prospectus, 30 December 2025). A document must establish effective ownership and control, not merely describe a transaction as a 'stake' or 'nationalisation.' Apply Resolution date (A3) and the Resolution window convention.

Download all forecasts (JSON)