OpenAI offers Washington a 5% stake, which is a slightly strange thing to offer

OpenAI is in early talks to give the US federal government a 5% equity stake in the company, worth roughly $42.6 billion at its March valuation of $852 billion. The vehicle would be a public wealth fund modelled on the Alaska Permanent Fund. Sam Altman has pitched it in person to Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent, and White House AI adviser David Sacks. Altman also wants Anthropic, Google and Meta to contribute matching 5% stakes. Anthropic denied on the same day that any such conversation was happening.¹²

This is a slightly strange arrangement and I want to walk through why it is that way.

What was actually reported. The Financial Times had the story first on 2 July; CNBC, Axios and the Guardian followed the same afternoon.¹²³ Every version describes the discussions as "early stage" with "no formal offer made." The White House has not commented. The vehicle proposed is a national analogue to the Alaska Permanent Fund, the constitutional trust established in 1976 that converts state oil royalties into resident dividends and now manages roughly $80 billion.⁴ The 2023 Alaska dividend was $1,312 per eligible resident. Altman has met with Alaska Gov. Mike Dunleavy and former Sen. Dan Sullivan, presumably because when you are borrowing someone's institutional design it is polite to ask.

Why now. OpenAI converted to a Delaware public benefit corporation earlier this year, with the original nonprofit now holding approximately $40 billion in cash rather than equity in the operating company.⁵ Bankers have been selected for an IPO targeted as early as late 2026.⁶ In parallel, the Department of Defense triggered a formal review of GPT-5.6 deployment in mid-June, delaying commercial rollout by several weeks. Anthropic's Fable 5 model has been suspended pending DoD clearance since roughly the same window.⁷ Bipartisan Senate discussions of mandatory pre-deployment reviews for frontier models were underway in late June.⁸ The proposal to hand Washington 5% of the cap table arrived about two weeks after those reviews began to bite. This is not, I think, a coincidence.

That is what 5% of OpenAI is nominally worth at the last round. It is also roughly the amount of cash the nonprofit received in the for-profit conversion, which is a coincidence but a suggestive one — the price of the political concession is being pitched at approximately one nonprofit-conversion's worth of equity.

The structural read. There are two ways to describe this proposal. One is Altman's framing: a public wealth fund that shares AI's upside with citizens, modelled on a widely admired state institution. The other is that OpenAI is offering to dilute itself by 5% in exchange for regulatory alignment with the primary regulator ahead of an IPO. Both descriptions are true. The interesting question is which one governs the term sheet.

The Alaska Permanent Fund analogy does not really survive contact with the facts. Alaska's fund holds a diversified portfolio of public-market securities funded by resource royalties collected from many operators. What Altman is proposing is a concentrated equity position in four private companies, taken directly from the companies themselves, in a structure where the shareholder is also the regulator. Norway's Government Pension Fund Global, at $1.7 trillion the world's largest sovereign wealth fund, explicitly caps its ownership at 10% of any single company specifically to avoid becoming a controlling shareholder.⁹ A US federal fund starting at 5% concentrated in four regulated frontier labs would be the opposite of the diversified public-vehicle model it is invoking.

Anthropic's denial is the tell. For the wealth fund to work as pitched, all four labs need to contribute simultaneously. If OpenAI gives 5% and its peers do not, OpenAI is diluted relative to competitors that retain the same regulatory exposure at lower cost. Anthropic denied any stake conversations on the same day the FT story ran.² Google and Meta have not confirmed anything. Four-party coordination absent congressional compulsion is a low base-rate outcome, and Brookings has previously noted that the executive branch does not have independent authority to accept and manage private equity stakes without an act of Congress.⁹ So the operational path to actually consummating this — new legislation, matching participation from three other companies, a governance structure that resolves the regulator-as-shareholder conflict — is long and narrow.

Which suggests the proposal is doing most of its work as a public proposal, not as an executable transaction. The signalling is: OpenAI is willing to be aligned with the US state. Whether the wealth fund ever exists is secondary to whether Washington believes Altman is willing to build it.

What this is a case of. This is the "AI safety as market position" frame executed at the political layer rather than the product layer. The usual moves inside that frame are Responsible Scaling Policy updates, defence-market carve-outs, safety research releases timed for enterprise sales cycles. This is a step upstream: converting political risk into political alignment by giving the regulator an equity interest in your success. The cost is 5% dilution at a peak private valuation. The benefit, if it works, is a federal counterparty that profits from OpenAI's growth rather than being incentivised to constrain it.

The AI performativity frame — the idea that scale of spend and scale of gesture make AI materially influential regardless of what it delivers — also fits here. Offering the US government $42.6 billion in equity is a gesture of a size that structures the political conversation whether or not it is ever paid.

The IPO problem. If any version of this happens before OpenAI files, the S-1 risk factors are going to be interesting to read. Standard IPO disclosure language assumes regulatory adversarialism: here are the agencies that could hurt us, here is how. A cap table where the federal government owns 5% and also regulates the company scrambles that structure. Precedent exists — the US Treasury held large equity stakes in banks after 2008 and in General Motors after 2009 — but those were rescue arrangements, structured to be temporary and unwound at the earliest opportunity. This would be the opposite: a permanent equity position taken pre-IPO, with the shareholder retaining rulemaking authority indefinitely.

Institutional investors pricing an OpenAI IPO would need to decide whether the government's 5% is an alignment premium (regulatory approvals move faster) or a governance discount (a veto player sits on the cap table). At 34x revenue on a $25bn annualised run-rate, there is not much room in the multiple to absorb the second interpretation.

What to watch. Three things. First, whether Anthropic, Google or Meta makes any public statement that softens the collective-action problem. Second, whether any senator, Warner is the natural candidate, given Altman's outreach, introduces enabling legislation, because without that the fund cannot exist. Third, whether DoD reviews on GPT-5.6 and Fable 5 clear in the next four weeks. If they do, we will know what the 5% was for.

Footnotes