The Stake

The Paranoidist | Issue #23 By Paul Morin | July 12, 2026

Ten days ago the Financial Times reported, and CNBC, CNN, and TIME confirmed, that the chief executive of OpenAI has proposed giving the United States government roughly 5 percent of his company. The stake would sit in a sovereign-wealth-fund-style vehicle modeled on the Alaska Permanent Fund, the arrangement Alaska has used since 1982 to convert oil royalties into an annual dividend check for every resident. At the valuation OpenAI reached in its record March funding round, 852 billion dollars, the stake would be worth roughly 42.6 billion dollars, and the proposal as described envisions the other leading American AI developers ceding similar slices to the same vehicle. The reception has been the most improbable coalition in Washington. The President confirmed last month that he has discussed arrangements in which, as he put it, the American public "essentially becomes a partner" with the companies. A socialist senator has a bill on the table that would take ten times as much. And the best-known executive in the industry is volunteering his own shares. When the populist right, the progressive left, and Silicon Valley converge on the same instrument, the commentary writes itself: the upside of artificial intelligence, finally, shared.

I want to concede the appeal before I doubt it, because the appeal is real. The concentration of AI's gains is the observable state of the world, and no existing mechanism does much about it. The instrument itself is real and tested: Alaska has run its fund for five decades, Norway's holds on the order of two trillion dollars of the world's equities, and by one Forbes commentary's tally more than twenty American states already operate public wealth funds of their own. If artificial intelligence is what its builders say it is, then 5 percent of the industry, volunteered, would be the largest broad-based wealth-sharing gesture in American history. The thing that keeps me up is not that the public would own a piece of the machines. It is the direction the payment runs, and what a payment in that direction quietly purchases from the one institution whose whole value to the public is its freedom to say no.

Issue #22, "The Excess," ended on a builder selling capacity it swore it could never have enough of, and on the question that has run under this entire series: whether the demand justifying the AI build-out is genuine end demand or a small cohort of firms financing one another's conviction. This issue moves that question one level up, from the balance sheet to the statehouse, because the newest buyer being recruited into the structure is the government itself. Call the two readings of the offer what they are. A dividend is the public's share of an upside it already underwrites: if AI is the new oil, the fund is the royalty arriving, collected as of right by an owner, the way Alaska collects on oil pumped from land Alaskans own. A retainer is a standing fee paid to keep a professional engaged and favorably disposed: the same 5 percent, offered by the regulated to the regulator, purchasing goodwill from the referee whose job description is the willingness to blow the whistle. The same stake is a royalty or a fee, and the entire question is which way the payment runs.

Start with the comfort story at full strength, because on its own terms it is the best version of an idea serious people have wanted for years. The gains from this technology flow to a narrow set of shareholders and employees while the disruption lands on everyone, and the standard policy toolkit, antitrust, privacy, safety rules, touches conduct but never ownership. A public fund touches ownership. It is not a novelty and not socialism by another name: Norway's fund owns slivers of most of the world's large companies and pays for a welfare state; Alaska's pays a dividend to every resident of a deeply conservative state; the sovereign funds of the Gulf compound quietly across decades. The design question is solved. What has always been missing in the American case is the asset, because the federal government does not own an oil field. The proposal's elegant move is to treat frontier AI as the oil field.

And the politics, for once, point the same way from both ends. The President has spent his second term converting subsidies into equity and export permissions into revenue, and has said openly that he likes the idea of the public becoming a partner in the AI companies. Senator Bernie Sanders introduced legislation in June, the American AI Sovereign Wealth Fund Act, that would impose a one-time tax of 50 percent of the stock of systemically important AI companies and deposit the shares in exactly this kind of fund. OpenAI has been building toward the idea in public for over a year: its April policy paper proposed a public wealth fund whose returns could be distributed directly to citizens, including those who own no financial assets, and reporting indicates its chief executive first raised the concept with the administration in early 2025. On the comfort story's assumptions, this is that rarest of things, a genuinely bipartisan mechanism for sharing a genuinely transformative technology, offered voluntarily, before the fight that would otherwise force it.

Now read the same offer from the other side, and start where the reporting itself starts. The Financial Times, which broke the story, describes the proposal's purpose, per two people familiar with the talks, as a way to keep relations with the administration warm and to "address political blowback." That is not my characterization; it is the stated logic of the offer as reported. And it changes the instrument's nature entirely, because a dividend and a retainer are distinguished by exactly one thing: the direction of the obligation. A dividend flows from an asset the owner already holds; Alaska's residents are paid because Alaska owns the oil under its land, and the royalty is a property right that exists whether or not the oil companies feel generous. Nobody in Juneau describes the Permanent Fund as a way to secure good relations with the producers. Here the ownership would be created by the payment: the public is offered a stake it never held, by the firms the public's government regulates, as a gesture whose reported purpose is the goodwill it generates.

The contrast with the Sanders bill makes the point cleanly; the two proposals are not versions of each other despite arriving at the same vehicle. Under the tax, the state takes; the companies owe nothing afterward and are owed nothing. Under the donation, the founder gives; and a gift, in every context humans have studied, creates a relationship. The same fund arrives by two roads, and the road decides who holds the leverage when the fund exists. In any other regulated industry, a substantial voluntary transfer from the supervised to the supervisor is not called visionary; it is the precise thing the ethics rules exist to prevent, and routing this one through the public's benefit makes it novel, not different. The talks are described as early conversations, no terms exist, any deal would likely require an act of Congress, OpenAI declined to comment, and the White House did not respond. All of that is true, and none of it is the point. The point is what the offer is for, and the reporting has already told us.

The state-as-shareholder is not hypothetical; it is the operating template of the current administration, and the proposal lands into it. Last August the government converted unpaid CHIPS Act grants and a defense program award into a 9.9 percent stake in Intel, roughly 8.9 billion dollars of equity, part of a total federal investment of 11.1 billion; the Commerce Secretary marked the occasion by saying Intel was excited "to welcome the United States of America as a shareholder." Weeks earlier, Nvidia and AMD agreed to pay the government 15 percent of their chip-sale revenue from China in exchange for permission to sell there. The government has taken stakes in critical-minerals producers, holds a golden share in US Steel with veto power over specific corporate decisions, and has, by CNBC's description, invested in quantum computing companies including IBM. The Center for Strategic and International Studies tallies roughly ten billion dollars of federal funds committed to equity positions so far and calls it a marked shift in American industrial policy. The AI proposal is not an outlier arriving from nowhere. It is the largest possible extension of a pattern already priced.

What makes the AI case different from Intel, and sharper, is the specific set of hats the prospective shareholder already wears. In June the President signed an executive order establishing a framework under which AI companies voluntarily provide the government access to their models for up to thirty days before release, for national security review. Within weeks, OpenAI announced it was restricting the release of its newest model, GPT-5.6, at the administration's request, and said publicly that it does not believe that kind of government access process should become the long-term default. Days after that, a presidential directive instructed the national security agencies to speed up their adoption of the most advanced models from multiple vendors. Hold those three facts together. The same government is now the gatekeeper who can hold a frontier model at the door, the anchor customer being ordered to buy more of it, and, under the proposal, the shareholder whose portfolio rises when the company it gates and buys from does well. Three roles, one balance sheet. There is one more mirror worth noticing: OpenAI's record March round was co-led by a fund backed by Abu Dhabi's sovereign wealth apparatus, so the company already knows what it is to have a state on the cap table. The proposal would add the state that writes its rules.

The deepest problem is that the arrangement's advocates do not deny the alignment of interests; they advertise it. If the state benefits when AI prospers, the argument runs, the state will let AI prosper. That is the feature. But a referee whose retirement account is denominated in one team's success does not need to be corrupt for the game to be corrupted; he needs only to be human, and the players need only to know what he holds. The institution cannot see the problem because the problem is the design working as intended. Capture, here, would not be a malfunction. It would be the product.

What lifts this above one company's Washington strategy is that three things are arriving in the same window and pointing the same way. The first is the arc this series has been tracing since spring. The ratepayer is already financing the grid for data center load that counts as firm when it sets the bill and stays optional for the tenant (Issue #21). At least one of the largest builders has conceded, in the politest available language, that it built more capacity than it needs (Issue #22). The offer of upside to the public arrives, in other words, at the exact moment the insiders' own signals have begun to soften, and the 42.6 billion dollar headline is priced off the very valuation whose demand assumptions the last two issues put in question. The public is being offered equity in the bet in the same season the bettors have started to hedge it, and a gift whose value depends on the giver's most contested number is a gift worth appraising before it is framed.

The second is the administration's broader equity-era statecraft, which binds the government's political fortunes to asset prices more tightly every month. The same week the proposal broke, the administration launched its branded investment accounts from the Oval Office and the President remotely rang the opening bells at both major exchanges, celebrating the market's records as proof his agenda is working, while Federal Reserve data show the top 1 percent of earners hold half the country's equity wealth. A government whose political scoreboard is the index, holding stakes in the companies that drive the index, acquires a compounding interest in never letting the number fall. That is not a partisan observation; it would be true of any administration so positioned. It means the state's stake would not sit passively in a drawer. It would sit inside every future decision about whether to gate a model, police a merger, or let a correction run.

The third is sequence. June was the month the government demonstrated, on the public record, that it can hold a frontier model at the door for a month, and the month a directive made it the industry's accelerating anchor customer. July is the month the industry's most prominent firm proposed making that same government a shareholder. Whether anyone intended the sequence is unknowable and beside the point; institutions read sequences, and every AI lab in the country has now read this one. The available lesson is that regulatory exposure can be managed with equity, and once one lab learns it, declining to learn it becomes expensive for the rest. Which is, quietly, how a voluntary gesture becomes a standing toll.

The honest forecast comes in three parts, and only the first is close to mechanical. Near term, nothing passes. The talks stay conceptual, no bill advances quickly, and the proposal does its work unpassed, because the goodwill is generated by the gesture rather than the statute: the headlines describing generosity, the meetings the offer convenes, the option on the relationship that stays open. Expect more instruments from the same voluntary family in the meantime, pledges, access agreements, gestures, the same species I flagged when seven companies signed the Ratepayer Protection Pledge, because the voluntary instrument delivers the relationship without the vote. Intentions, then as now, are not firewalls, but they photograph well.

The medium term turns on two questions with no dates on them. The first is whether Congress ever codifies anything, and on what terms, because the terms are the entire difference between Norway and capture: voting shares or non-voting, board seats or none, a statutorily independent manager or a politically appointed one, a firewall between the fund and the agencies that gate models and award contracts, or a hallway. The second is whether any of the other laboratories the reporting names publicly declines, and what happens to a decliner. Whether any of them would participate is genuinely unclear; at least one, according to a source cited by Reuters, has had no discussions with the administration about a stake at all. The treatment gap between participants and refusers, in pre-release reviews, in procurement, in export decisions, is the natural experiment that will eventually reveal what the 5 percent actually buys, and it is the single most informative thing to watch.

The counter-move, when it comes, will not be a clean rejection, and this is the part to brace for. If the fund dies in Congress, expect the relationship to be delivered by other means: warrants and revenue shares on the chip-deal template, procurement preferences, expanded voluntary access, equity-adjacent instruments that never require a vote. The base case is not a public wealth fund and not a scandal. It is a thickening lattice of financial ties between the referee and the players, assembled piece by piece, each piece individually defensible, none of it ever put to the public whose name is on all of it. So watch the gauges. Watch for actual bill text, the Sanders Act's movement, and any drafted governance terms, because voting rights and manager independence are where the binary resolves. Watch which laboratories sign, which refuse on the record, and how each is treated afterward in reviews and contracts. Watch the count of days models spend in pre-release review, by lab. Watch the federal equity portfolio's growth from its current committed base of roughly ten billion dollars. And watch OpenAI's own listing documents when they come, because a company gearing up to go public as soon as this year must describe its government relationships as risk factors, under oath of accuracy, and that filing may be the most honest paragraph anyone writes about what the stake is for.

Four areas of board exposure, mapped against the retainer reading rather than the dividend.

AI developers, their investors, and anyone holding pre-IPO exposure. This is where the reading cuts first, and it cuts both ways. Participate, and you carry a state shareholder whose other hands hold your release gate and your largest contracts, plus dilution booked as political necessity. Decline, and you carry the refusal, in an environment where the government has demonstrated it can hold a model at the door and where your competitors' goodwill is now capitalized. Either way, some part of your valuation has quietly become a judgment about political proximity, and proximity is revocable: it does not survive an administration change or a falling-out, and it can reprice overnight in a way a technology moat cannot. The question for the board and the investment committee is whether your models treat regulatory goodwill as a durable asset or a revocable license, and what happens to the multiple on the day the license is revoked.

Boards in every strategically designated sector, not just AI. The template migrates, and the tally says it is migrating fast: chips, critical minerals, steel, quantum, with defense contractors publicly floated as next. Any company that can be described as critical to national security may eventually face the invitation, and the worst moment to decide your answer is when the call comes, under deadline, with a subsidy or an approval hanging on the tone of your reply. The question for the board is whether you have a policy, adopted in advance and owned at the board level, for what you would concede if Washington proposed equity participation, revenue sharing, or a golden share, what you would refuse, and who is authorized to say either word.

Institutional allocators, pensions, and endowments. A federal vehicle holding private stakes in the largest companies ever to approach the public markets changes the political economy of the listing calendar you invest into, and the state's interest in high AI marks reaches your portfolio through policy whether or not you hold a single AI name. You may also, in time, find the government sitting beside you on cap tables as a co-shareholder that simultaneously writes the rules, gates the product, and buys the output, a counterparty your governance frameworks were not built to score. The question for the investment committee is whether your risk and governance analysis prices a co-shareholder with sovereign powers, and whether your AI exposure assumes valuations that political proximity currently supports and political rupture would not.

Enterprises buying AI, which is to say nearly everyone. Vendor selection is acquiring a political dimension. A supplier's models can be held at the door for a month by its newest prospective shareholder, which from your side of the contract is an outage with a flag on it; state-partnered and independent suppliers may diverge in access, export treatment, and continuity risk in ways no benchmark captures. The question for the technology and risk committees is whether vendor due diligence now includes the supplier's political balance sheet alongside its technical one, and whether continuity planning covers a critical vendor whose next release is a matter of state.

The dividend may simply be right, and overdue. The concentration of AI's gains is real, the fund mechanism is proven across five decades and multiple political cultures, and the perfect arrangement that shares the upside without touching the referee does not exist and never will. On that reading, insisting on an immaculate mechanism is a fastidious way of ensuring the public gets nothing, the offer is exactly what it claims to be, and the correct response to the largest voluntary wealth-sharing gesture in American history is to take it and write good rules, not to psychoanalyze it.

Ownership may improve governance rather than corrupt it. Norway's fund holds stakes in most of the world's large companies, and Norway has not stopped regulating them; more than twenty American states run funds without dissolving their oversight. Design can wall off capture: non-voting shares, a statutorily independent manager, information barriers between the fund and the gating agencies. A state with a financial stake also has better information and more standing than one peering in from outside, and the arms-length alternative has not obviously produced wiser technology governance anywhere it has been tried.

The proposal may be a trial balloon that dies quietly. The conversations are described as early, no terms exist, the company declined to comment, the White House did not respond, and any real version needs an act of Congress that the current one may never take up. The modal outcome of most trial balloons is nothing. If so, I am treating a lobbying gambit as a constitutional turn, and the retainer worry, however sound in principle, attaches to an instrument that will never exist.

And the capture may already be priced, in which case formalization is transparency. Influence already flows between the industry and the state through lobbying, personnel, procurement, and access, none of it visible on any balance sheet. An on-the-books equity stake is disclosed, auditable, and repealable in a way the existing influence economy is not. If the relationship exists regardless, the version with a line item may be the more honest one, and objecting to the visible form while tolerating the invisible one gets the risk exactly backwards.

The DeepStrategy.ai signature method requires sorting risk, which is quantifiable, from uncertainty, which is not, at every major analysis. The numbers here are risk. The 852 billion dollar valuation and the roughly 42.6 billion dollar implied stake, the roughly ten billion dollars of federal equity committed to date and the terms of each position, the Intel stake's size and price, the 15 percent China revenue shares, the count of days each lab's models spend in pre-release review, the status of the Sanders bill and any successor text, the lobbying disclosures, and, in time, the risk-factor language in the industry's listing documents can all be tracked, charted, and put on a board dashboard tomorrow. They will tell you the size and direction of the move.

What will not resolve to a figure is what the move means. Whether a shareholder state regulates its holdings impartially is an uncertainty, because it will be tested only case by case, each case contested, and no case ever cleanly attributable to the stake. Whether the same 5 percent is a dividend or a retainer is an uncertainty, because the payment looks identical either way and only the subsequent pattern of decisions, spread over years, reveals the direction of the obligation. Whether an arrangement like this, once created, can ever be unwound is an uncertainty of the one-way-door kind: funds are built to be permanent, and permanence is precisely what makes the terms matter more than the intent. The precise figure is the announcement; a 5 percent stake worth roughly 42.6 billion dollars is a fact anyone can compute. Whether that payment would buy the public its share of the future or buy the industry its regulator is the part that will not reduce to a probability, and the two answers point in opposite directions from the same stake.

The institution that consumes the analytical process as preparation for multiple futures has what the forecast cannot provide: adaptability. If the fund advances, the headlines will tally what the government would hold: the percentage, the billions, the dividend per citizen, and the coverage will call it the public finally getting its share. The architecture asks the quieter question the title has been carrying all along. The stake worth discussing is not the one the government would hold. It is the one on the table: the referee's freedom to say no, the single public asset that has no market price, offered up in exchange for a slice of an upside the public was already underwriting through its grid, its savings, and its patience. You do not have to doubt anyone's motives to see the shape. You only have to ask what the payment is for, and the reporting has already answered.

The boards best positioned for this are not the ones reading the offer as generosity or as scandal. They are the ones asking, of every counterparty and every regulator relationship in their own plans, which way the payments run, what each gesture of goodwill purchases and from whom, and pricing the answer before the pattern of decisions makes it public.

A gift from the regulated to the regulator. A dividend the public never had the standing to demand. And a referee being offered shares in the game it was hired to call.

The Paranoidist publishes weekly, with flash issues when events warrant.

If this changed how you think about one thing, consider subscribing. If it didn't, tell me what I'm missing.

Paul Morin is the founder of DeepStrategy.ai, author of Uncertainty: When Risk Is Not Enough (a guide to decision-making when probabilities fail), and publisher of The Paranoidist, BoardroomRadar, and ScenarioWatch. He has spent more than three decades in entrepreneurship, finance, risk management, and insurance, which is why he worries about the things that keep other people awake at night.

Researched, written, and edited in collaboration with Claude by Anthropic. Anthropic is itself one of the leading AI laboratories this issue describes: the reporting on the proposal names it among the developers envisioned as ceding a stake, and a source cited by Reuters says Anthropic and the administration have had no such discussions. The reader should weigh this analysis knowing it was produced with the tools of a company that holds a position in the game the issue examines.

All information reflects publicly available reporting verified as of Sunday, July 12, 2026. The status of the OpenAI talks, any congressional action, other laboratories' responses, the federal equity tally, and the valuation figures can change without notice; re-run a news scan for developments on the proposal and re-verify the moving figures immediately before posting.