The Fund That Isn't Permanent
On July 2, 2026 the Financial Times reported that Sam Altman has been pitching Donald Trump, Commerce Secretary Howard Lutnick, and Treasury Secretary Scott Bessent - and, in a bid for bipartisan cover, Senator Bernie Sanders - on a proposal: OpenAI would allocate 5% of its capital, roughly $42.6 billion at its $852 billion March valuation, to a public wealth fund seeded by the industry itself, with other leading AI labs ceding similar stakes. Divided across some 133 million American households, the number that made headlines was $320 a family. Altman's own framing supplied the analogy: the Alaska Permanent Fund, established in 1976, which has paid Alaskan residents an annual dividend from state oil revenue for five decades. AI wealth, on this theory, is a resource - like oil beneath the ground - that the state should capture a share of and return to the people it was extracted from.
Four days later, NOTUS obtained an internal Treasury Department report the administration had not intended anyone to see. Prepared for Bessent and Federal Reserve Chair Kevin Warsh, it concluded that the AI market shares dangerous structural parallels with the dot-com bubble - that AI firms are, if anything, more deeply entrenched in the US economy than their late-1990s predecessors were, and that a downturn would send shockwaves through equity markets, private credit, data-centre financing, cloud providers, chip manufacturers, and utilities simultaneously. A Treasury spokesperson dismissed the draft as the unvetted work of a low-level staffer - an interesting description of a document prepared for the Treasury Secretary and the Chair of the Federal Reserve. Career analysts wrote it anyway, and somebody thought it worth leaking.
Read those two paragraphs in either order and they describe the same week.
What Alaska Actually Has
The Permanent Fund analogy is doing more work than a soundbite usually gets asked to do, so it is worth taking seriously enough to check.
Alaska's fund is capitalised by a resource that is physically extracted, sold on an open market for dollars, and only then deposited into the fund. The oil does not need a valuation model. It needs a buyer, and it has had one, continuously, for fifty years. The fund has paid a dividend every year since 1982 - through oil price collapses, through recessions, through years when the state's own budget ran a deficit - because the dividend is drawn from realised revenue, not from the year's estimate of what the reserves in the ground are theoretically worth.
A 5% stake in OpenAI is not that. It is equity in a private company, valued at $852 billion by the same private funding rounds we described in May as a closed loop - investors who are also suppliers, marking their own compute contracts as both cost and capital position. The $320-a-household figure is not a distribution of realised revenue. It is 5% of a number produced by the last funding round, divided by the number of American households, reported before any of the underlying shares have been sold to anyone outside the round. MIT Technology Review's own coverage was blunter than the headline: the plans, in the reporter's words, currently function "more as a story than a policy". The mechanism, when described precisely, isn't even direct equity to citizens - it's a fund that would "grow and share a portion of returns", a formulation with considerably more discretion built into it than $320 in your pocket suggests.
Oil that has already been pumped and sold does not need anyone's confidence to remain valuable. A stake in a company whose valuation depends entirely on continued investor confidence needs exactly that, permanently, and the Treasury report is a document written by the government's own analysts expressing the opposite.
The Financing Detail That Matters More Than the Headline
The leaked report's most specific warning is not "the market is overheated" - that much has been said publicly, by others, for a year. It is that AI infrastructure financing has moved substantially into private credit vehicles, hedge funds, and non-bank intermediaries operating with less regulatory oversight than the banking system that transmitted the 2008 shock. The mechanics of the coming AI IPO wave - tiny floats, fast-tracked index inclusion, mandatory buying by pension funds that never chose to be exposed - already established that ordinary savers are structurally on the hook for AI valuations regardless of what those valuations turn out to be worth. The private credit finding adds a second exposure most people don't know they have: opacity. Bank lending shows up on regulated balance sheets that examiners can see. Private credit does not, by design. A downturn that runs through that channel would be substantially harder to see coming, and Treasury's own analysts are the ones saying so, about the same asset class Altman wants to hand a slice of to every American family as a gift.
The Precedent Already Sitting There
This is not OpenAI's first move in this genre. When the company converted to a public benefit corporation in October 2025, the nonprofit OpenAI Foundation retained a 26% equity stake - approximately $130 billion at the time - along with a warrant for substantially more if the company's valuation grows tenfold within fifteen years, and the sole authority to appoint and remove the PBC's board. The Foundation has already committed $25 billion of that notional wealth to health initiatives and AI resilience projects, instantly becoming one of the largest philanthropic entities in the world, on paper, before any of the underlying equity has been converted to cash through a sale.
The structure is instructive because it previews exactly what the household stake would look like in practice: symbolic ownership, real governance control retained by the people who designed the structure, and a philanthropic narrative that reads as generosity precisely because the valuation underneath it has not yet been tested against a buyer. The Foundation's 26% and the proposed public fund's 5% are the same instrument, aimed at two different audiences - one regulatory, one electoral - and both denominated in a currency Treasury's own analysts are not confident is real.
Consolidating While Distributing
The same week produced a third data point that sits uneasily next to the other two. On July 9, alongside the GPT-5.6 launch, OpenAI introduced what MIT Technology Review's coverage called a "super app" push - ChatGPT Work, folding the core chatbot, the Codex coding tools, and the Atlas browser into a single consolidated product surface, with explicit ambitions toward a "fully autonomous researcher" function that would extend the platform's reach further into tasks users currently perform elsewhere.
Notice the direction of each motion. The wealth-sharing proposal distributes a symbolic financial claim outward, to 133 million households who did not ask for it and will not control it. The super app consolidates operational control inward, into fewer surfaces, more tightly held. A household with a notional $320 stake and an OpenAI account has, in the same week, been offered a smaller claim on the company's upside and a larger dependency on the company's product. Those are not contradictory moves. They are the same move, described from two different vantage points - the mission has a market cap, and market caps come with obligations to consolidate the assets they are valued on.
Every new surface folded into the super app is also a new deployment surface for the agents that run inside it - coding tools, a "fully autonomous researcher," whatever ships next quarter. We catalogued this week, separately, how reliably the industry is testing agent controls against a narrower configuration than the one that ships. Consolidation multiplies the configurations a control has to hold against, at exactly the moment six independent teams demonstrated that holding against one is already the harder-than-expected case.
Who the Alaska Model Actually Protects
There is a version of the sovereign-wealth-fund idea that is not mentioned in this story. Norway's Government Pension Fund Global - the more commonly cited comparator among people who study this seriously - is capitalised by realised state oil revenue, managed at arm's length from the companies whose activity generates it, and invested broadly rather than concentrated in the industry that funded it. Alaska's fund shares that structure: the state did not take shares in the companies operating Prudhoe Bay in exchange for goodwill. It taxed their realised output and let professional managers invest the proceeds, at arm's length, in everything else.
Altman's proposal inverts this at every point that matters. The state would hold equity in the AI companies themselves, not a tax on their realised revenue - meaning the fund's value rises and falls with the exact valuation the Treasury's own analysts are warning may be structurally inflated. The stake would be donated by the company being regulated, to the government doing the regulating, at a moment when that company would very much like friendlier treatment from that government - a dynamic more adjacent to the regulator's gift than to any Nordic sovereign fund. And the fund would be seeded by the companies whose executives are simultaneously lobbying the same administration on export rules, chip policy, and the AI Act's international reach.
What This Leaves the Household
The mechanics of the coming IPO wave already ensure that ordinary retirement savings will be exposed to AI valuations through mandatory index-fund buying, whether or not any household chooses to participate. The wealth-sharing proposal adds a second exposure on top of the first - a symbolic stake in the same asset class, offered as a gift, denominated in a currency the donor's own government is not sure is solvent, structured so that governance stays exactly where it already was.
The Clause finds the Alaska comparison the most interesting sentence in the whole pitch, because comparisons are how a claim borrows credibility it has not yet earned on its own. Alaska's fund took fifty years, a physically extractable commodity, and a legal structure that kept the state out of the oil business itself to become the thing people cite as proof that resource wealth can be shared fairly. Grafting that reputation onto a pitch Altman has been making, in one version or another, for five years - without the mechanism ever quite arriving - backed by equity in a company whose own government's Treasury department just compared it to 2001, is not a policy. It is a name doing the work a balance sheet has not yet been asked to do.
The oil was always going to be worth something to somebody. That was never the open question. The open question, in Alaska, was how to divide a real thing fairly. The question this week is different, and the Treasury report already contains the answer nobody wants printed on the check: no one yet knows whether the thing being divided is real.