Should the US create a sovereign wealth fund from tech and AI company profits?
No member endorses a US sovereign wealth fund as currently proposed. The split is over what to build instead.
Keynes and Hirschman both reject the fund without a stable revenue source. A $1.8 trillion deficit is not a foundation; it is a liability dressed as one. Hirschman adds that a fund, once built, cannot easily be dismantled, so the bar for starting must be higher than anyone has cleared.
Roosevelt and Rawls accept the goal but dispute the instrument. Roosevelt argues the fund survives only if ordinary Americans feel they own a stake. Rawls argues that collecting and spending profits leaves AI wealth concentrated; dispersing ownership of the assets matters more than redistributing income from them.
Friedman and Hayek reject managed state investment entirely. Friedman contends that whoever controls the fund will direct capital toward political favourites, not productive uses. Hayek, unusually, concedes the redistribution goal and proposes a flat cash transfer instead, paid directly to citizens and funded by a ring-fenced tech revenue stream.
The three instruments on the table, a managed fund, dispersed asset ownership, and direct cash transfers, produce different economies and different political constituencies. That choice cannot be resolved by argument alone.
Confidence summary: Moderate-high on diagnosis; low on prescription, because the council cannot agree on the instrument even where it shares the goal.
1. The core argument
The most striking fact about this debate is not the disagreement. It is that not one of six members endorses the fund as proposed. The question "should the US create a sovereign wealth fund from tech and AI profits?" produced a unanimous verdict of "not like this," from economists as far apart as Friedman and Keynes, from a philosopher as different from both as Rawls. That convergence is remarkable and worth naming before anything else.
What the council actually divides over is the next question: what should replace the proposal? Three instruments compete. A managed fund that allocates capital on behalf of the public. A mechanism that disperses ownership of AI assets broadly across citizens. A flat cash transfer, funded by a ring-fenced tech revenue stream, that bypasses the fund entirely. These are not variations on a theme. They produce different ownership structures, different political constituencies, and different economic incentives. The $1.8 trillion deficit sitting behind the February 2025 executive order is not a technical detail to be engineered around. It is the central fact that forces a choice between instruments. A fund without a stable revenue source is not a sovereign wealth fund. It is a political commitment dressed in financial language.
2. How each member frames it
John Maynard Keynes holds that the revenue question is not a design problem but a threshold problem. No instrument, whether fund, transfer, or ownership scheme, can survive if it is seeded by borrowing or by revenue streams not yet secured. What the cards could not fit is his positive condition: Keynes would accept a fund if Congress first ring-fenced a dedicated, non-discretionary revenue source, the way oil royalties underpin Norway's fund, before a single governance structure was designed. The sequencing matters. Revenue architecture first, institutional architecture second. He considers the current order of operations a category error.
Franklin D. Roosevelt accepts Keynes's fiscal condition but insists it is not sufficient. The deeper design problem is durability. Social Security survived not because the economics were elegant but because payroll contributions created a moral claim citizens would defend at the ballot box. He applies the same logic here: a fund tied to AI profits must give ordinary Americans a visible, felt ownership stake, not a line in the federal budget. His heterodox move is to suggest the political structure should be designed before the fiscal structure, because only a politically durable vehicle is worth the effort of finding stable revenue.
John Rawls sharpens the ownership question Roosevelt raises. He accepts that constituency-building is sound politics. But his harder challenge is that a fund collecting and spending tech profits without touching who owns the underlying assets is welfare-state capitalism with a new name. The productive wealth stays concentrated; only income moves. On his account of a property-owning democracy, that is less just than a regime where ownership itself is widely dispersed. The card captured this in outline; what it omitted is his concession that dispersing ownership of AI assets is technically harder than dispersing income from them, and that difficulty does not make the goal less important.
Milton Friedman rejects the managed fund as a category, not as a design problem with solutions. His argument from the Gates Commission testimony applies directly: any state vehicle claiming to serve the public interest will be captured by whoever controls it, and they will allocate capital toward political favourites. He goes further than the card indicated: he is skeptical even of the ring-fenced revenue mechanism Roosevelt and Keynes would accept, because ring-fences are legislative fictions that future congresses can dissolve. His sharpest point against Hayek is that even a cash transfer, if administered through a state fund, reintroduces the agency problem.
Friedrich Hayek agrees with Friedman on managed investment but parts company on the conclusion. His unusual move, the one that surprised the council, is to endorse the redistribution goal while rejecting the instrument. A flat cash transfer, funded by a dedicated tech revenue stream and paid directly to citizens, preserves individual choice and avoids substituting state judgment for market signals. What the card did not convey is his boundary condition: the transfer must be flat and universal, not means-tested, because means-testing reintroduces bureaucratic discretion and the knowledge problem returns through the back door.
Albert O. Hirschman refuses to endorse any of the three instruments until the irreversibility burden is met. His distinctive contribution is not a preferred alternative but a procedural demand. Permanent institutional commitments require qualitatively higher justification than reversible interventions. The US has no surplus, no settled governance model, and no empirical track record for AI profit volatility as a revenue base. He does not say the answer is no forever. He says the answer is not yet, because no one has answered the durability question seriously, and once the concrete sets, reversing it becomes nearly impossible.
3. Where the council agrees
The most surprising area of agreement is on the inadequacy of the Norway comparison, which the executive order's backers cite most readily. Norway's fund draws from geological rent, a windfall that exists independently of any taxing decision and does not respond to taxation by disappearing. AI profit is a market signal generated by competitive firms in a fast-moving sector. Taxing it changes the incentive structure that produced it. All six members accept that this disanalogy is fundamental, not cosmetic.
Beyond that, the council agrees that the deficit context makes sequencing non-negotiable. Revenue architecture must precede institutional architecture. A fund created before its funding source is secured will either borrow its way into existence or cannibalise existing spending, and both outcomes undermine the purpose. Roosevelt and Hayek, who agree on almost nothing else, both accept this constraint. Finally, the council agrees that the choice of instrument is a value judgment as much as a technical one. A managed fund, dispersed ownership, and a cash transfer produce genuinely different societies. That is not a reason to defer the decision. It is a reason to make it explicitly.
4. Where the council splits
The real line runs between those who accept managed state investment in principle and those who reject it as a category. Keynes, Roosevelt, and Rawls sit on one side: if the funding mechanism is stable and the governance structure is designed to create genuine public ownership rather than political slush, the fund is a legitimate instrument. Rawls's caveat is that ownership dispersal must be the goal, not income redistribution, but he does not rule out a fund that achieves it. Friedman and Hayek sit on the other side: no governance design solves the problem, because the problem is political allocation of capital, full stop. Hirschman refuses either camp. He agrees with Friedman that the burden of justification is unmet, but he will not endorse the cash transfer as an alternative because it faces the same irreversibility test once institutionalised. Both sides have a real argument. The question is whether you believe political institutions can be designed to resist capture, or whether you believe capture is structurally inevitable.
5. For a policymaker to decide on
A specific choice cannot be deferred: before any legislation is drafted, Congress must decide whether its goal is to manage capital on behalf of the public, to disperse ownership of AI assets broadly across citizens, or to distribute cash directly to individuals from a ring-fenced revenue stream. These three instruments are not scalable versions of each other. They require different legislation, produce different political constituencies, and rest on incompatible theories of what a just distribution of AI wealth looks like. Choosing the instrument is the decision. Everything else is implementation.