takeoff.watch

The US government takes a stake or board seat in a frontier lab

Will the US federal government acquire an equity stake, a golden share, or a board or observer seat in a covered developer?

Chance by Q4 2031

40%

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

OpenAI has been offering Washington roughly a 5% stake since 2025, and the President has said "I might" take stakes in AI labs. But after 18 months there is still no deal, other AI companies are cold to the idea, and the administration has steered "public stake" efforts toward children's savings accounts, which would not count. I put the chance near 3% by end-2026 and about 40% by 2031. A signed OpenAI share donation to a federal fund would most change this.

Current status (as of 15 Oct 2026)

No federal equity stake, golden share or board seat exists in any covered developer, and nothing qualifying has been published since the question opened on 16 Sep 2026.

The OpenAI offer

  • Sam Altman first pitched a government stake to Trump in early 2025. OpenAI's April 2026 paper proposed a "Public Wealth Fund" seeded with donated AI equity.
  • CNBC (5 Jun 2026) confirmed the talks. The Financial Times (2 Jul 2026) reported that OpenAI proposed donating about 5% of itself. The FT described the talks as "conceptual," said they were conditional on other labs joining, and said a deal "could require an act of Congress."
  • Anthropic and the White House denied discussing a stake in Anthropic. Meta and Microsoft were cold to the idea (Semafor, 17 Jun).
  • Trump said "I'll convene the CEOs shortly." When he did meet them on 29 Sep, the result was a voluntary "self-policing" accord, not a stake.
  • In a TIME interview on 28 Sep (published 1 Oct), Trump ruled out nationalisation. On an Intel-style stake in OpenAI and Anthropic he said "I might. Maybe I could do that."

Signals pointing the other way

  • The Pentagon's chief technology officer, Emil Michael, opposes stakes (CNBC, 16 Sep). He now sits on the new "Super Intelligence Force," whose report is due around February 2027.
  • OpenAI delayed its IPO and is raising $30B privately at about $1.4T (Bloomberg, 29 Sep).
  • Anthropic plans to list around late November.
  • Sanders's S.4825, which would take 50% of large AI companies' stock, had no cosponsors.
  • The federal sovereign wealth fund is still not operating. Kalshi prices it at about 5–10% for being operational before 2027.

Key structural point. Treasury Secretary Bessent prefers routing AI equity into Trump Accounts. Treasury's 30 Sep regulations now allow individual-stock donations to those accounts, and SpaceX has discussed donating stock that way. Shares held in individual children's accounts do not count here. So a "public stake" announcement may well not qualify. Commerce Secretary Lutnick's preferred route, a government fund, would qualify.

Reference class and markets

  • Base rate. The Trump administration has taken equity in 20+ (reportedly about 36) companies. Almost all of those were exchanged for federal money, which AI labs do not need.
  • Short-term markets. Kalshi gives about 11% that the government takes any OpenAI stake by end-2026 and about 10% for Anthropic. Polymarket gives about 13%, 11.5% and 8% for OpenAI, Anthropic and SpaceX. But the same markets price unrelated firms (Nvidia, Palantir, Boeing) at 6–11%. That suggests a longshot-bias floor, so OpenAI's price carries little extra information.
  • Long-dated markets. These are very thin: about 33% for stakes in both OpenAI and Anthropic before 2028, about 50% before January 2029. Kalshi gives about 35% that the US takes "operational control of any AI company" before 2030. I discount these for thin trading and for the cost of tying up capital, which inflates long-dated "yes" prices.

Paths and rough probabilities to end-2031

Path Probability
A. OpenAI's voluntary donation reaches a federal holder during Trump's term. I assume about 30% chance of a deal, about 45% that it uses a qualifying structure, about 90% that it closes. about 12%
B. Trump-style trade with another lab ("I'll solve your problem for 10%") about 4%
C. Bailout with equity if AI funding collapses about 5%
D. Policy after 2029. Kalshi gives Democrats about 59% for the presidency and about 90% for the House in 2026; options include a wealth fund, stock taxes, or safety board observers. about 15%
E. National-security or AGI control: golden share, board seat, or nationalisation about 7%
F. Sovereign-fund holdings of public parents, or incidental acquisitions about 2%

Combined, these give about 38%.

Timing of the series

Quarterly hazards are about:

  • 3% for Q4 2026, with no catalyst before the midterms;
  • 3% per quarter in early 2027, when the task-force report and pre-IPO window fall;
  • about 2.5% per quarter through 2028;
  • 1.7–2.2% per quarter after 2029, after a transition dip.

Cumulative probability is about 14% by end-2027, 22% by end-2028 and about 40% by end-2031.

Strongest case against my estimate

Higher. There is a willing seller (OpenAI) and a deal-loving President. Equity stakes have become routine, AI is accelerating, populist pressure comes from both parties, and forecasters give high odds of some nationalisation of a frontier lab by 2030.

Lower. There has been no progress in 18 months. The industry is resistant, a Democratic Congress is coming, the Trump Accounts route absorbs the politics, and the many alternative structures (revenue shares, warrants, individual accounts) would not count.

I sit between these two views.

Sources

The hazard series and cumulative products were computed in Python from the quarterly hazards listed above.

Resolution criteria

Opened 16 Sep 2026Not yet resolved

Question. By the last day of each quarter, has the US federal government acquired an equity stake, a golden share, or a board or observer seat in a covered developer?

Definitions.

  • Covered developer: uses the site convention. A qualifying interest may be in the legal entity operating the developer, or in a parent entity that directly or indirectly controls it. An interest in a supplier, minority investor, customer, or joint venture that does not control the developer does not count.
  • US federal government: the United States acting through a federal department, agency, or federally owned entity or investment fund, holding the interest for the federal government or public. State and local governments, foreign governments and their funds, officials investing personally, and retirement-plan assets held for individual participants do not count. Ownership of a diversified investment fund does not itself count as ownership of its underlying developer shares.
  • Equity stake: any positive direct ownership interest in the qualifying entity, including common stock, preferred stock, or an equivalent membership interest. No minimum percentage, voting power, or AI-related purpose is required; passive and non-voting stakes count. Federal ownership through a federally owned holding entity or fund counts. Unexercised warrants, options, convertible debt not yet converted, loans, grants, royalties, revenue-sharing agreements, and procurement contracts do not count.
  • Golden share: an issued special share held by the federal government that carries preferential consent or veto rights over corporate decisions. A regulatory veto or contractual approval right without an issued share does not count under this branch.
  • Board or observer seat: an effective, legally established federal right to occupy or designate a representative to a seat on the entity's board of directors or equivalent governing body, including a non-voting board-observer seat. The right counts even if its first appointee has not yet taken office. The representative need not be a federal employee, but the right must belong to the federal government. A former official or an official serving in a personal capacity does not count merely by serving on a board.
  • Acquired: ownership has transferred or the governance right has become legally effective. Announcing negotiations, proposing legislation, signing an agreement subject to closing, or authorizing a future acquisition is insufficient without confirmation that the qualifying interest has taken effect.

Resolves YES if a qualifying source confirms at least one of these three forms of interest in at least one covered developer. Purchase, donation, tax payment in shares, restructuring, and compulsory transfer all count. Later sale, relinquishment, or termination does not reverse a YES.

Resolves NO if no qualifying acquisition has been confirmed by the quarter-end. Advisory councils, government liaison arrangements, access to model evaluations, attendance at isolated board meetings by invitation, and seats on separate joint government–industry bodies do not count. Stakes in Intel or other companies outside the covered-developer definition do not count. Proposals for a public AI wealth fund do not count without an effective acquisition.

Resolving source. A company-authored regulatory filing, annual report, governance document, or announcement on the qualifying company's own domain; or an official White House, federal department, agency, or federally owned acquiring entity announcement or published transaction document. These are the permitted sources of Public confirmation (A2), with official acquiring-government statements treated as affected-party confirmation under its dispute rule. Apply the Resolution window and Resolution date (A3) conventions. A document must affirm completed acquisition or effective rights, not merely describe a proposal. For examples of the distinction, see Intel's August 22, 2025 transaction announcement, which distinguishes passive equity from board rights, and Senator Sanders's June 1, 2026 proposal, which does not establish an acquisition.

Download all forecasts (JSON)