Europe would love its own OpenAI. But in AI, the winner is whoever can compute the most, not whoever has the best ideas.

Every few months, the call goes up for a European OpenAI, a sovereign model out of the EU, financed with billions that get announced with great ceremony. Behind it sits one assumption: with enough will and enough money, Europe catches up. That assumption misjudges what kind of race this is.

How capable an AI model becomes depends in large part on how much compute went into training it. More of it usually means a stronger model, and the whole industry organises itself around that fact. The currency of this race isn’t good ideas. It’s chips, electricity and capital, in quantities that only a handful of companies can put up at all.

And those quantities are concentrated. In a single year, the four largest US firms alone spend far more on AI infrastructure than the EU holds out for its entire AI strategy. On top of that, the decisive hardware is scarce. The AI boom has triggered a shortage of the fastest memory chips, prices have risen several-fold within a year, and supply is spoken for years ahead, mostly to the same American buyers. You can’t simply buy your way in here.

Suppose the EU financed a champion of its own anyway, with public money. What it would create is a state-backed competitor in a capital race against firms spending several times its budget, dependent on exactly the American chips and data centres it was meant to free us from. Little about it would be European: the headquarters, the flag in the name. The compute would still come from the same source, the independence would stay on paper.

This is no verdict on European engineers or European research, both are excellent. It’s a statement about orders of magnitude. A race decided by who can buy the most of a scarce resource is not won by the side that starts with less of it and has less money to buy more.

It doesn’t follow that Europe holds nothing. It follows that copying OpenAI is the wrong plan. Europe’s weight lies where it is already indispensable: in the one company in the world that builds the machines for the most advanced chips, and in power over a market of more than 400 million people. Anyone who wants to sell into it has to play by the local rules. Setting those rules is Europe’s real lever, not a race for compute it cannot win.