That is what each American household would receive, in equity, if Sam Altman’s proposal to hand the US government a 5% stake in OpenAI were distributed evenly across American families.
Based on recent funding rounds, a 5% stake in OpenAI would represent tens of billions of dollars in paper equity.
Put those two data points together, and the political problem for AI companies is not abstract.
A public equity fund reframes that legal exposure as a settled political question.
A collective equity fund is compensation only in the sense that averaging a debt across 340 million people makes any individual creditor’s claim disappear.
The number that emerged from a leaked negotiation this month was $320. That is what each American household would receive, in equity, if Sam Altman’s proposal to hand the US government a 5% stake in OpenAI were distributed evenly across American families. Not per year. Not compounding. A one-time paper stake, sitting in a fund somebody else manages, in a private company nobody outside a small circle of institutional investors can currently buy shares in directly.
Framed as generosity, this is the most talked-about wealth-sharing proposal in American technology policy right now.
Framed as arithmetic, it is close to a rounding error on a household’s monthly grocery bill.
The distance between those two framings is where the actual story lives.
The proposal, and what it is actually proposing
Reports have emerged that Altman has been in talks with the Trump administration about giving the US government a 5% equity stake in OpenAI. The pitch, according to people close to those conversations, is that the American public helped train these models — through the books, movies, images, essays, and code that were scraped into the training data — and deserves a share of what the models produce.
Based on recent funding rounds, a 5% stake in OpenAI would represent tens of billions of dollars in paper equity.
Distributed across American households, this translates to approximately $320 per household in paper equity, a calculation that has circulated widely in recent analyses of the proposal.
That is the top-line number. It is also where the coherent part of the proposal ends.
Five years of the same idea, dressed differently
This is not the first time Altman has floated something like this. He has previously proposed that large companies pay a portion of their market value into a citizens’ fund, with proceeds distributed to Americans.
That version was more radical. It was also more specific. It named a mechanism, a rate, and a distribution channel.
In April, OpenAI itself published a policy document titled Industrial Policy for the Intelligence Age that laid out a narrower version of the same theme — the idea that AI-generated economic gains should be shared, with the government as a partner rather than a regulator.
Senator Bernie Sanders, from a very different political corner, has proposed legislation to create a $7 trillion AI sovereign wealth fund, giving Americans a 50% stake in the top AI companies rather than 5%.
Five years. Multiple versions. Wildly different numbers. Zero implementation.
That last fact is the one worth sitting with.
Why $320 is the wrong number to argue about
Critics of the proposal have latched onto the size of the household payout. $320 is not going to change anyone’s life. It will not offset a single month of childcare, will not cover a car repair in most of the country, will not pay for a semester of community college.
But arguing about whether $320 is enough misses what the proposal actually does.
It is not a redistribution mechanism. It is a narrative.
The purpose of putting a specific dollar figure on every American household’s paper stake in OpenAI is to make the AI buildout feel participatory — to convert public anxiety about job displacement, energy consumption, and creative theft into a story where ordinary people are, technically, shareholders. A 5% stake sounds like ownership. In practice, if the vehicle is structured as a passive fund with no voting rights and no distribution mechanism for years or decades, it is closer to a symbolic gesture with an accountant attached.
The Alaska Permanent Fund, which Altman’s proposal is repeatedly compared to, is the useful counter-case. Set up in the 1970s to distribute oil royalties to Alaska residents, it now pays out real annual dividends to residents. The mechanism is specific. The oil is a physical, finite resource. The state constitution requires it.
OpenAI is not oil. There is no constitutional requirement. There is no distribution schedule. There is, so far, a conversation.
What the polling actually shows
The political appeal of a wealth-sharing narrative becomes clearer when placed next to the underlying data on public sentiment.
A Bentley-Gallup survey on AI attitudes found that a majority of Americans do not trust companies to use AI responsibly. Concern about AI’s increasing role in daily life outweighs excitement for many Americans. The expectation that AI will improve individual lives in the near term remains limited.
Local resistance is even sharper. Recent Gallup polling on data center construction found that Americans broadly oppose data centers being built in their own areas, citing water use, electricity strain, and noise.
Put those two data points together, and the political problem for AI companies is not abstract. It is a public that is skeptical of the technology, distrustful of the companies building it, and physically resistant to the infrastructure required to run it.
A $320 paper stake does not solve that problem. But it changes the sentence structure of the argument. It moves the conversation from they are taking things from us to we all own a small piece of this.
That is worth something. Just not $42.6 billion in equity, most likely.
Photo by Tom Fisk on Pexels
The compensation-for-training-data argument
The subtler version of the proposal, and the one that has genuine legal and moral weight, is the framing that a public equity stake would function as belated compensation to the writers, artists, photographers, coders, and musicians whose work was used to train the models without payment or consent.
This framing matters because the alternative — resolving the training data question through the courts — has been going badly for AI companies, or slowly, depending on the case. Book authors, news organisations, image libraries, and record labels are all in various stages of litigation over whether scraping copyrighted material to train commercial models constitutes fair use.
A public equity fund reframes that legal exposure as a settled political question. If Americans, collectively, own 5% of OpenAI, then the training data debt has, in some sense, been paid. Not to the specific author whose novel was scraped, but to the abstract public that includes her.
The problem is that this is not how compensation works in any other domain. If a factory pollutes a river, the settlement does not go to every American household in a paper stake. It goes to the specific people harmed, in proportion to the harm.
A collective equity fund is compensation only in the sense that averaging a debt across 340 million people makes any individual creditor’s claim disappear.
Why the Trump administration is a plausible partner for this
The counterintuitive part of the current moment is that a Republican administration is the one actually engaging with what sounds like an aggressive state-ownership proposal. In previous political eras, a 5% federal stake in a private technology company would have been treated as socialism-adjacent by the political right.
The Trump administration’s approach to strategic technology has been different. It favours direct deals — equity, revenue-sharing, export controls tied to specific arrangements — over regulation. A 5% government stake in OpenAI fits that pattern. It is not a rule. It is a transaction.
For Altman, the appeal is that a bilateral deal with the White House is faster, more predictable, and more durable across news cycles than trying to shape legislation through Congress. It also gives OpenAI something its competitors do not have: an explicit alignment with US strategic interests, which matters for export licenses, defence contracts, and the ongoing question of which American AI companies get treated as national champions.
The cap table of Anthropic’s recent $65B raise already looked more like industrial policy than a venture round. OpenAI handing 5% to the federal government would make that pattern explicit.
The valuation problem underneath the proposal
There is a quieter question buried in the mechanics of the $320 figure, which is whether the valuation it is calculated from is a stable number.
OpenAI’s most recent funding rounds have valued the company at figures that would put it among the most valuable private companies in history. The $122B round it closed in March 2026 — the raise that set the $852B valuation the household figure is calculated from — was part of a broader pattern in which foundation model companies are raising capital at valuations that require enormous, sustained revenue growth to justify.
Those valuations are also being propped up by the ongoing collapse in inference costs. The 280-fold drop in the cost of running a top-tier AI query between late 2022 and late 2024 makes the economics of running a foundation model business genuinely difficult, because the same drop applies to competitors.
If OpenAI’s valuation halves — a scenario nobody at the negotiating table wants to plan for, but which is well within the range of what has happened to comparably-hyped private companies — then the household stake drops to $160. If the company IPOs at a lower valuation, as some analysts covering the recent IPO speculation have suggested is possible, the number moves again.
The $320 figure is not fixed. It is a snapshot of a private valuation on one specific date.
What a real version of this would look like
The gap between the current proposal and a functional wealth-sharing mechanism is not ideological. It is structural.
A real version of this policy would need to answer several questions the current proposal does not.
Who holds the shares? A sovereign wealth fund, a trust, a Treasury account? Each has different governance and tax implications.
What are the voting rights? Passive shares in a company whose board is already controversial is different from active shares that give the public a seat at the table.
When do distributions occur? Alaska pays annually. This proposal has no schedule.
What happens if OpenAI is acquired, restructured, or goes public? The mechanics of the stake surviving those events are non-trivial.
Does the same rule apply to Anthropic, Google DeepMind, Meta’s AI division, xAI? If not, why is OpenAI special? If yes, what is the legal authority to require it?
None of these questions have public answers yet. Which is itself the answer to the question of how serious the proposal is.
The pattern worth noticing
Zoom out from the specific numbers, and a pattern emerges that has less to do with OpenAI than with a recurring feature of the AI policy conversation.
Proposals that sound redistributive — universal basic income funded by AI productivity gains, citizen dividends from foundation model companies, sovereign wealth funds seeded by big tech equity — have been circulating for the better part of a decade. Almost none have been implemented anywhere. The ones that have been implemented (Alaska’s oil fund, Norway’s sovereign wealth fund) were built on physical, extractive resources where the state had clear jurisdiction and the flows of money were legible.
AI-generated wealth is not that. It is generated across borders, using inputs (training data, compute, human labour) that are diffuse and contested. Any wealth-sharing mechanism would need to decide who counts as a contributor, how contribution is measured, and how ongoing value is attributed.
The current proposal skips those questions by making the answer simple: everyone gets an equal share, calculated on one company’s valuation as it stood on a single day. That is a clean number. It is not the same thing as a plan.
