OpenAI wants to hand Washington 5% of itself. Read what the government would actually own.

The number that travelled this week was 5 percent — the stake in OpenAI that the company has reportedly floated handing to the United States government — and its companion figure, roughly 42.6 billion dollars, which is what 5 percent of an 852-billion-dollar company works out to. Both numbers are doing a lot of quiet work, and both deserve to be interrogated rather than repeated. Because a government equity stake is not a donation, and it is not a tax, and it is not the subsidy running the other way that Washington usually hands to industries it wants to keep. It is a position. Someone ends up long something. The interesting question, as with any position, is who is exposed, to what, and what breaks if the story underneath it turns out to be wrong.

The reported shape of the proposal, first, because the mechanism is the story. According to the Financial Times, Sam Altman has been discussing — directly with President Trump, Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent, and separately with Senator Bernie Sanders — an arrangement in which OpenAI would grant the federal government an equity stake of around 5 percent, with the idea that other American AI companies would be asked to cede similar stakes into a sovereign-wealth-fund vehicle modelled on the Alaska Permanent Fund. The talks are described as conceptual. They may require an act of Congress. No other company has agreed; Anthropic has reportedly said it has not even discussed the idea. Trump called it “interesting” and said officials would look into it. That is the whole of it, at this stage: a framing, a precedent invoked, and a lot of things that have not happened yet.

What a stake is, as distinct from what it's called

Altman’s framing is generous and civic: give the public a financial interest in the upside of AI, the way Alaska gave its residents a share of the oil. It is a good line, and it is worth translating out of the dialect, because the translation changes what you are looking at. An equity stake is a claim on future profits and a seat, however quiet, at the ownership table. When the owner is the state, the state acquires two things at once: a financial interest that rises and falls with the company’s fortunes, and — this is the part the Alaska analogy is built to make you not think about — a financial interest in the outcomes of its own decisions about that company. Alaska’s fund owns a diversified basket of assets it does not regulate. The United States would own a slice of the single most consequential company in an industry it is, at this exact moment, writing the rules for.

Hold those two facts next to each other, because the entire risk of the idea lives in the space between them. This is the same government that has spent the spring gating access to frontier models — approving the rollout of the most advanced systems customer by customer — that has used export controls as an on-off switch on specific models, and that is being asked, through the courts and a multistate subpoena, to scrutinise how OpenAI treats its users and whether its conversion to a for-profit structure passes muster. Give that government 5 percent of the company, and every one of those decisions acquires a second dimension. A regulator weighing whether to approve a model, block a deal, or press an investigation is now, at the margin, weighing the value of an asset it holds. That is not a corruption story. It is a structural one, and structure is more durable than any individual’s good intentions.

The United States would own a slice of the single most consequential company in an industry it is, at this exact moment, writing the rules for. — On the conflict the Alaska analogy is built to hide

Why it's cheap for OpenAI, which is the tell

Start from the trader’s reflex: when someone offers to give away something valuable, price what they get back. Five percent of OpenAI is, on paper, an enormous concession — 42.6 billion dollars is not a rounding error even at this altitude. But no cash leaves the building. The company issues equity, dilutes its existing holders by a sliver, and hands the paper to the one counterparty whose goodwill is worth more to it than almost any amount of money: its own regulator. Read as a financial transaction it looks like generosity. Read as risk management it looks like one of the cheapest insurance policies available to a company in OpenAI’s specific predicament, which is that its greatest threat is no longer a competitor’s model but the state’s posture toward it.

Consider what the political risk actually is, and how neatly a shared equity interest hedges it. OpenAI faces an antitrust environment sharpening around the largest AI firms, a subpoena from dozens of state attorneys general, a contested restructuring from non-profit to for-profit that regulators must bless, and a growing public backlash against the data-centre build-out its future depends on. Every one of those is a case where it helps, enormously, if the government sitting across the table has already decided it is better off with OpenAI thriving than struggling. A shareholder is not a neutral party. That is precisely the point. The company is not buying a subsidy with this stake. It is attempting to convert its most dangerous adversary into a stakeholder, and doing it with paper rather than cash — which, if you are OpenAI, is the trade of the year.

There is precedent close to hand, and it cuts both ways. Last year the government took roughly 10 percent of Intel — the state stepping in as a shareholder in a strategic chipmaker it wanted to keep alive. But the direction of that deal is the mirror image of this one. Intel was the weaker party accepting capital and validation from a stronger state. Here, the company is the one initiating, offering, and structuring the arrangement — the regulated proposing to make its regulator an owner. When the party with the most to gain designs the mechanism, it is worth reading the mechanism very carefully.

The valuation is doing the heavy lifting

Now the other number, the 42.6 billion, because it rests on a foundation softer than its precision suggests. Five percent is only worth 42.6 billion if OpenAI is worth 852 billion, and that 852 billion is not a market price. It is a private mark — negotiated in funding rounds among a small set of insiders who are all, structurally, long the number, because the value of their own holdings, their ability to recruit, and the credibility of their next raise all depend on it holding. It has never been discovered in the one place that discovers prices, which is a public market with sellers in it as well as buyers. OpenAI has, by its own reported choice, delayed the listing that would test the figure rather than risk the market marking it lower.

So the stake being offered is a percentage of a valuation that is itself an assertion. If the public interest in AI’s upside is meant to be the justification, it is worth noticing that the upside is being measured in a currency the seller prints. A 5 percent stake in a company worth 852 billion is worth 42.6 billion. A 5 percent stake in the same company, if a public market decided it was worth half that — which is a real range for an unlisted firm burning cash against enormous forward promises — is worth 21 billion, and the gift the public was told it received quietly halves while nobody sold a share. The headline figure is not a fact about what Washington would own. It is a fact about what OpenAI would like the number to be, extended to a new and unusually credulous buyer.

Stack the moving parts and the position resolves into something less flattering than the press framing:

• The concession costs OpenAI no cash — it is dilution, paid in paper the company values itself.
• The buyer is the regulator, so the stake doubles as insurance against the antitrust, restructuring and investigation risks that are the company’s real exposure.
• The 42.6-billion valuation of the stake is a private mark that has never survived a public bid, and shrinks with any honest repricing.
• The Alaska framing imports the language of a diversified, arm’s-length fund onto a concentrated position in a company the owner regulates — the two situations share a structure only in the brochure.

The downside no one is pricing

Here is the exposure the civic framing skips. Spread these stakes across OpenAI, Anthropic, Google, Meta, and the chipmakers, as the proposal reportedly imagines, and the government does not become a neutral steward of a diversified basket. It becomes long the entire AI trade — concentrated, correlated, and, unlike Alaska’s oil royalties, invested in the outcomes of its own regulatory and national-security choices. If the AI build-out delivers, that is a windfall the political class will be very glad to have bought. If it disappoints — if the capex proves to have outrun the revenue, if a model cycle turns, if the depreciation on all that hardware arrives faster than the returns — the taxpayer is now the believer holding the position into the drawdown, and the referee is now financially invested in the score staying high. Concentration is the risk that compounds quietly and arrives all at once, and a state that owns a piece of the thing it is supposed to police has arranged its incentives so that the moment it most needs to act against the industry is the moment doing so costs it the most.

None of which means the underlying idea — that the public should share in the gains of a technology built on public research, public data and public risk — is wrong. It is a serious idea and deserves a serious instrument. The point is narrower and it is a markets point: this particular instrument, an equity stake initiated by the company and priced off a valuation only the company can vouch for, hands the government upside and the company insurance while quietly binding the regulator to the fortunes of the regulated. Whether that is a good trade for the public depends entirely on a question the announcement is careful not to ask, which is what happens if the thesis is wrong. A stake is a bet. The people being enrolled in this one have not been shown the downside case, and in my experience the deals that hide the downside case are the ones where somebody has already read it.