OpenAI Just Put a Price on the Federal Gate. The Bid Is $42.6 Billion.

The Financial Times broke it Thursday: OpenAI has been pitching the Trump administration on a 5 percent equity donation into a US sovereign wealth fund structured after the Alaska Permanent Fund. Sam Altman ran the concept through Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent. The proposed vehicle is not just for OpenAI; the framework asks Anthropic, Google, Meta, and xAI to each cede 5 percent into the same fund. At OpenAI's March post-money mark of $852 billion, the check is $42.6 billion.

For six months TF has been tracking a federal gate around frontier models: the June 12 Fable 5 pull, the NCD preview list around GPT-5.6 Sol, the White House voluntary release framework. That gate has always had a compliance cost. On Thursday it picked up a price tag.

The Math

Read the two rows on Intel and OpenAI side by side. Washington paid Intel $8.9 billion in cash last August for 9.9 percent of a $90 billion chipmaker, plus a five-year warrant on an additional 5 percent. OpenAI is proposing to hand over 5 percent worth $42.6 billion for zero cash. Per point of equity, the OpenAI offer is roughly 9.5x the Intel transaction on a valuation basis (Intel priced at about $900M per point, OpenAI at $8.5B per point). Even after you strip out the fact that the Intel deal was cash-out and this one is a donation, the ratio tells you which company is more scared of the administration and which one Washington is more nervous about upsetting.

Why This Number, Why Now

The FT scoop landed six days after the GPT-5.6 Sol preview list was published under the White House voluntary predeployment framework, and 20 days after Anthropic's own Fable 5 got pulled off the shelf for the same review process. The federal gate we wrote up on Sol and the buyable frontier took the flagship of the largest closed lab in the world off-market for two months. When the compliance surface is a green light for your top SKU, the marginal cost of a bad relationship with the administration is a multiple of what an equity donation would run.

Altman's pitch was not new. He floated the sovereign-fund idea to the White House in early 2025, well before GPT-5.6, before the NCD gate, before the Anthropic S-1. What is new is the price. In early 2025 OpenAI was marked at roughly $300 billion. 5 percent of that was $15 billion. Nine months later the mark is $852 billion and the same 5 percent is $42.6 billion, a $27 billion increase in the size of the check without a single new term negotiated. The valuation ran faster than the pitch did, which is why the number in the FT looks so large now.

The Match Ask Nobody Wants

The proposal is structured as an industry ask, not a single-lab ask. That is the part the other four names on the list are going to spend the weekend reading. On the same day the FT ran the story, a person familiar with Anthropic's side told CNBC that the company has not discussed a government stake with the White House. The White House, Google, and Meta all declined comment. Nobody wants to be the second lab to agree; nobody wants to be the last lab to refuse.

Do the same math on the other four. Anthropic filed the confidential S-1 on June 1 at a $965 billion mark; 5 percent is $48.3 billion. Google Alphabet trades at roughly $3.3 trillion; a 5 percent AI-carveout on the DeepMind and Google Cloud AI franchises is harder to price cleanly but easily runs into the $100B range once you unwind shared services. Meta at $2.1 trillion market cap; a 5 percent AI-specific carveout is at least $50B and probably more once Llama, Reality Labs, and the Meta neocloud push get priced in. xAI at $200 billion post-money; 5 percent is $10 billion. The total ask across the five names sits somewhere north of $250 billion of equity into a fund that has not yet been chartered by Congress.

That is the tell. $250 billion of frontier-lab equity dwarfs anything the Alaska Permanent Fund vehicle Altman keeps citing has ever managed (Alaska is at about $80 billion of assets today). It also dwarfs the $12 billion Congress appropriated for the CHIPS Act. The proposal is not a program the Treasury can spin up out of existing authority. It needs a new statute and a new fund. That timeline runs long past any of the IPO windows the labs are currently pointing at.

What This Does to the Anthropic S-1

Anthropic has 32 days into a confidential filing window that the SEC clock will run for at least three more months before the first amended draft goes public. Kira covered the June convergence when Trump and Sanders arrived at the same policy from opposite ends of the spectrum; the FT scoop is the concrete instantiation of that convergence. Anthropic can now truthfully write in the risk-factors section that a leading US AI lab is currently in talks with Treasury and Commerce about ceding a 5 percent equity stake to the federal government, and that comparable transactions could be requested of the company in the future. That paragraph did not exist as an actionable risk factor a week ago.

For a filer, the paragraph cuts both ways. On the upside, it lets the underwriters argue that the discount for regulatory risk is already priced into the roadshow. On the downside, it hands the pricing committee an excuse to shave 10 to 15 percent off the target mark. Anthropic's bankers were probably marking to $965B until Wednesday. As of Thursday, marking to somewhere in the $825 to $875B range is a defensible number, and a defensible number is what actually goes into an S-1 amendment.

What This Does to the Price Floor

A 5 percent donation is not an operating cost. It does not show up in cost of goods sold, does not compress gross margin, and does not change the marginal cost of an inference call. It is a one-time balance sheet event that reduces the residual equity claim of existing shareholders by 5 percent. On paper, it should not touch the inference price floor at all.

In practice, we think it puts a modest upward pressure on the frontier tier, and only on the frontier tier. Reason: the labs that would participate are the closed labs at the top of the ladder, and the labs that would not participate are the open-weights crews (Meituan, Alibaba, Mistral, and the smaller US open players). If the administration extracts a 5 percent tax on the closed frontier and imposes no equivalent cost on the open frontier, the effective marginal cost of running a US closed lab goes up (dilution is a real cost of capital) while the effective marginal cost of running an open lab does not. That widens the price gap between closed and open exactly at the moment LongCat-2.0 just topped OpenRouter on hardware US export controls cannot reach. The federal gate would be widening the door open-frontier crews walk through.

Our Take

The Altman offer is a defensive trade dressed up as a public benefit. If OpenAI can get 5 percent onto the government cap table for zero cash, the company converts a regulator into a stakeholder. Stakeholders do not pull SKUs off the market for voluntary safety reviews. Stakeholders do not slow-walk export licenses. Stakeholders renegotiate. That is a real product for OpenAI, and it is worth more to OpenAI than $42.6 billion of paper. From the OpenAI seat, this is a cheap trade.

The problem is what it does to everyone else at the table. Anthropic now has to decide whether the S-1 amendment carries a matching commitment (in which case the underwriters shave the mark) or does not (in which case the risk factor gets uglier and the bookrunner still shaves the mark). Google gets asked to carve out a US AI subsidiary that is separately capitalizable, which the antitrust division has been asking for on different grounds for two years. Meta and xAI get pulled into a conversation Zuckerberg and Musk have both spent years telling the market they were not going to participate in. Adjust for the political cost of refusing the ask, and the net is that a defensive OpenAI trade lands as an offensive weapon against the other four cap tables.

For builders and agents, the near-term signal is small. The equity gate does not change the SDK, the tokens per second, the SWE-Bench Pro score, or the AFTA receipt format. It does change the answer to two operating questions on a six to twelve month horizon. First, whether the closed frontier labs get their voluntary review windows shortened once they are on the cap table (probably yes, at the margin). Second, whether the open frontier crews accelerate their capture of the buyable ladder while the closed labs sit through the negotiation (also probably yes, at the margin). The two together compress the closed-model price premium a little further and buy Meituan, Alibaba, and the US open labs another quarter of runway to convert benchmark leads into procurement wins.

Three signposts in the next 60 days. One, whether Bessent's Treasury or Lutnick's Commerce actually publishes a term sheet, or whether the pitch stays a leak. Term sheet means the negotiation is real; leak-only means Altman was floating a trial balloon and the White House let it drift. Two, whether Anthropic files an S-1 amendment with a matching commitment before the confidential window closes. If it does, the match ask has traction. If it does not, the ask is unilateral and gets harder to enforce. Three, whether xAI ends up on the list at all. Musk has been publicly warm on the administration and publicly hostile on equity dilution; the resolution of those two positions is the tell for whether the sovereign fund lands as a bipartisan program or a Trump-era one that unwinds with the next administration.

The federal gate the industry has been engineering around since Fable 5 got pulled just picked up a dollar figure. It is $42.6 billion for one seat, north of $250 billion for the full table, and it is on offer from the largest closed frontier lab before Congress has drafted the statute that would receive it. That is not a policy proposal in the ordinary sense. It is a term sheet floated at the White House by a CEO trying to buy his way past the gate his own product is currently sitting behind. The next 60 days tell you whether Washington cashes the check.