According to a Wall Street Journal report picked up by Reuters, the Pentagon is reportedly in talks to lend around $5 billion to Fluidstack, a young computing company, in order to consolidate the American data centre supply chain. This is not a purchase contract: it is private infrastructure financing out of the defence budget.
We have been following the transformation of AI into heavy industry for months. This is the stage where the state stops being a customer and becomes a financier.
The difference between buying and financing
This point is not rhetorical. A state that buys computing is a customer like any other, subject to the same terms and the same prices. We saw this with the portal opened to three million personnel, where three suppliers share the market.
A state that lends is something else. It takes on risk, it acquires a claim, and above all it gains a capacity to influence the company's decisions: where to build, whom to sell to, according to which priorities.
It is an instrument that states use for the sectors they deem strategic — aerospace, energy, semiconductors. Its extension to computing says clearly which category AI has been placed in.
Why a young company rather than a giant
The choice is interesting. The dominant players do not need a public loan: they raise tens of billions on the markets, as we have documented on several occasions.
Financing a smaller player serves a different objective: creating capacity that does not belong to the same few. A state that depended entirely on three companies for its computing would find itself in the position we described regarding the company that refuses to sell: a position of weakness in the face of private decisions.
It is the same reasoning as an industrialist who finances an alternative supplier so as not to depend on the dominant one. We saw this with the agreement between Amazon and Qualcomm. Applied by a state, it becomes industrial policy.
This move brings the American model closer to the one it criticises. We reported a Chinese five-year plan worth $532 billion as an example of state coordination, by contrast with an American model resting on private initiative. A defence loan to a computing company is an instrument of industrial policy, whatever the label. The difference between the two systems is shrinking, and it is the constraint that is shrinking it.
The questions it raises
What conditions? A public loan generally comes with conditions: location of facilities, priority access, restrictions on customers. These conditions determine whether the operation amounts to support or to control.
What selection? Choosing one company rather than another is a decision that commits billions and that advantages one player in its market. The criteria should be public; they rarely are in defence files.
What risk? We document every week the uncertainty surrounding the real economics of this sector, from the absence of measured productivity gains to margins announced without audit. Lending five billion to a young company in this context is a bet, and it is the taxpayer who carries it.
What to take away
This information, if confirmed, marks a stage we saw coming without knowing what form it would take.
Computing is now treated as a national strategic resource, on the same footing as energy or critical components. That implies public financing, conditions, and a growing entanglement between corporate decisions and state decisions.
For Europe, this is useful information. We were asking ourselves whether you have to be the best to count, noting that a €388 million supercomputer was not playing in the same category. If the United States moves to direct financing, the question is no longer whether Europe should intervene, but with what instruments and at what scale.