What does a state own
when it owns AI?

The summer the instrument found agreement

On 18 June 2026, Senator Bernie Sanders introduced a bill transferring to the public 50% of the capital of the largest American artificial intelligence companies, through a one-time tax paid in shares and deposited into a sovereign wealth fund. According to its author's presentation, the fund would be worth around 7,000 billion dollars at current valuations, and an independent commission would exercise its voting rights.

Two weeks later, the Financial Times revealed that OpenAI proposes ceding around 5% of its capital to the federal government, close to 43 billion dollars at its March valuation, and suggests that its competitors contribute an equivalent share to a public vehicle inspired by the Alaska Permanent Fund. The President of the United States had expressed as early as June his interest in the idea of the public becoming a partner of these companies. The instrument, moreover, is no longer hypothetical: through other channels, the federal state already holds a stake of around ten per cent in Intel.

The unusual character of this convergence must be measured. A socialist senator, a Republican administration and the laboratories themselves ordinarily agree on nothing. They agree here on the instrument: a share of the capital. The disagreement now bears only on the proportion, five per cent offered against fifty demanded. When adversaries converge on the same instrument, it is most often because that instrument spares them all a more difficult question. It is that question that must be asked.

What capital confers

It would be wrong to treat the idea with condescension. It answers a well-founded concern, companies grown more powerful than many states, with the oldest instrument in the law: becoming their owner. And company law knows precisely what ownership confers. A right to profits. Votes in general meeting. Seats on the board. A right to information. At fifty per cent, with a commission exercising the voting rights, the state would block transactions, weigh on appointments, oppose disposals. Anyone who has sat on a board of directors knows these powers are not decorative.

But one must read what this right is. The shareholder, even a majority one, exercises its power within the corporate form: it deliberates on the interest of the company, in the language of the company. Every question that reaches a general meeting arrives already translated into the only idiom it knows, that of value. Capital is a remarkable instrument for asking the question it knows how to ask. It knows how to ask only one.

What capital does not transfer

Yet the singular power of these companies does not lie only in the decisions a board can take. It lies in what the systems carry within themselves, and earlier positions of this chamber have described it plane by plane. The norm of behaviour embedded in the models, those texts that determine what a system accepts, refuses or privileges when it answers hundreds of millions of people every day: it is written, revised and interpreted inside the company, through acts that pass through no assembly.

The library on which the model was trained, what was read, in which language, chosen by whom: it was constituted upstream, and no corporate organ deliberates on it. As for the very categories in which these systems are judged, they are elaborated elsewhere than in boardrooms.

None of this appears on the balance sheet. None of this is an asset that the accounting entry transfers with the share. A board of directors approves accounts, authorises transactions, dismisses executives; it does not draft the constitution of a model and does not sit on the choice of its corpus. These decisions remain, after the transfer as before, in the hands of those who already took them, and they do not become, by the mere fact of a shareholding, objects of deliberation for the shareholder. The shareholder state acquires a considerable right over the fruits of a power whose exercise is not transferred to it. It becomes the owner of the results of the norm. It does not become its author.

The displacement

And the operation does not leave the state where it found it. American commentators have already noted the conflict of interest of a regulator holding shares in what it regulates. The point is accurate, but it falls short of what is at stake. A conflict of interest, the law knows how to treat: it is declared, recused, sanctioned. Here, the interest would be voted by Congress, entered in the budget, and its dividend distributed to every citizen. An honest, public interest, spread across the entire body politic, is precisely the one no recusal procedure can reach. Norway, which the fund's promoters invoke as a model, has felt this weight for years with its own fund: one weighs with difficulty against what one lives on.

The overall picture then deserves to be faced. Europe regulates without owning: it writes obligations on systems whose norm and library are decided elsewhere. America is preparing to own without governing what matters: it takes a share in companies and gives itself, by the same act, a patrimonial reason no longer to constrain them. Two symmetrical answers, and symmetrically incomplete: each grasps the object with the only instrument its tradition hands it, the rule for one, the title for the other. Neither grasps the plane where the power is exercised.

Closing

The question posed by these companies is not first a question of sharing wealth, even if it borrows its form. It is a question of authority over what now takes part in the formation of judgement, that of institutions as well as that of citizens. To treat this question as a matter of ownership is to answer as an owner a question of jurisdiction; it is to acquire a revenue where an office should be held.

One does not buy a jurisdiction. One buys a dividend.