OpenAI says the public should get a stake in AI. Is that real?

The short answer: it is real enough to take seriously and not yet real enough to count on. The proposal gets the single most important thing right, a share of the assets rather than a tax on the profits, and leaves unanswered every detail that decides whether such a thing survives. The difference between those two outcomes has a history, and the history is worth ten minutes of your time.

What has actually been proposed

In July 2026 it became public that Sam Altman, the chief executive of OpenAI, had proposed donating 5 per cent of the company's equity to an American public wealth fund. At OpenAI's current valuation that stake is worth around 42 billion dollars. The idea is that other leading AI firms would contribute similar stakes, the fund would hold them on behalf of the public, and citizens would receive a dividend, the way every Alaskan has received a yearly cheque from the state's oil fund since 1982.

He is not alone. Senator Bernie Sanders has proposed something blunter: a one-time 50 per cent levy on the largest AI companies, paid in stock rather than cash. Three House Democrats have introduced a bill taxing AI companies' revenues to fund public employment. The debate about who should own the machines has, quietly and quite suddenly, become a real one.

That matters, because for most people the AI conversation is still about jobs: will the machine take mine, and when. The ownership question sounds abstract next to that fear. It is not. It is the answer to it. If the machines end up doing a large share of the work, then what a household lives on is decided by whether it holds a claim on what the machines earn. A wage stops when the job stops. A stake does not.

The test that separates the real thing from the gesture

Proposals like these have been made before, about oil, about land, about industry, and enough of them have succeeded and failed that the pattern is known. Four questions sort the survivors from the rest.

Is it a stake in the assets, or a tax on the flows?

A tax takes a slice of this year's profits, and stops the moment the political weather turns or the profits are moved elsewhere. A stake owns part of the machine itself, whatever it earns, wherever it earns it. On this test the Altman and Sanders proposals both pass, which is genuinely new. Equity, not revenue. The token-tax bill fails it: it meters the output and leaves the machine's title untouched.

Does it reach everyone, or pass through gatekeepers?

Alaska's fund mails a cheque to every resident. No application, no means test, no board deciding who deserves it. The moment a benefit is routed through institutions and eligibility rules, it acquires managers, and managers acquire discretion. The public wealth fund proposals gesture at a universal dividend, but none has yet committed to one on paper.

Can it survive the people who will one day want to raid it?

This is where most versions die, and it is the question on which the OpenAI proposal is weakest. Alaska's fund has paid out for four decades because it is written into the state constitution; when politicians tried to divert it, the design stopped them. A voluntary donation has no such armour. What a company gives to manage political pressure, it can renegotiate when the pressure fades. A gift is not a claim. The donor of a gift stays in charge of what it means. Nothing published so far says who would govern this fund, whether its shares would carry votes, or what stops a future Congress, or a future Altman, from quietly unwinding it.

Does it exist before the gains are distributed, or after?

Ownership hardens fast. Once the wealth of this build-out settles onto balance sheets, moving it requires taking it from someone, which is a different and much uglier politics than claiming it while it forms. On timing, all these proposals score well, for the simple reason that they exist now. This is the window. It is also the reason the subject deserves more public attention than it gets.

The honest reading

Be clear-eyed about motive. The 5 per cent offer surfaced while OpenAI was under acute political pressure, and 5 per cent is, among other things, an excellent price for legitimacy. A company offering the public a small stake in itself is also buying the public's acquiescence in the other 95. None of that makes the idea wrong. Alaska's fund was born of politics too. It does mean the details are everything: an unarmoured fund with a passive stake and no defined dividend would be the appearance of public ownership doing the work of preventing the reality.

And one number for scale. Five per cent of OpenAI is about 42 billion dollars. The capital spending of the large AI firms in 2026 alone runs to several hundred billion. The offer on the table is a real beginning and a small fraction of what is being built.

What to watch

Whether any proposal puts the fund's independence in law rather than in goodwill. Whether the dividend is universal and written down. Whether the shares vote, and who casts those votes. And whether it happens in the next few years, while the wealth is still forming, or becomes another study.

The public conversation about AI is a conversation about fear: jobs, layoffs, the machine coming for the work. The proposals described here are the first mainstream acknowledgement that there is a second conversation, the one that actually decides how this ends. Not what the machine can do. Who owns the machine?

---

Where the evidence lives. The four tests are developed in chapters 7 to 9 of The Horse Is Here to Stay: the case for stakes over taxes (chapter 7), the two funds that actually work and the design features that keep them alive (chapter 8) and the governance problem of who votes the public's shares (chapter 9). The running record of proposals as they appear is kept in the Stable.

One of a series of free guides answering the questions people are actually asking about AI and work. The book carries the evidence, the sources and the counter-arguments in full. These guides are published under a Creative Commons Attribution-ShareAlike licence: translate them, teach from them, quote them, build on them.