Europe Is the Third Pole That Cannot Act as One

Every time access to frontier AI is restricted, Europe starts talking about sovereignty again. Build our own models. Fund a champion. Protect the stack.

It is the wrong response to the right shock. Europe's problem was never the model. It is that Europe cannot act as one.

That sounds like a policy argument, and it is usually made as one. It is not. If you sit on a board or allocate capital in Europe, fragmentation is not something you are waiting for Brussels to fix. It is a standing condition you are underwriting in every AI bet you make, and it decides which European bets can work and which cannot, whatever the technology looks like.

The demonstration was blunt. In June 2026 the United States restricted its most capable models to US passport holders, and within days the restriction was in force for everyone, everywhere. It was reversed by the end of the month. The reversal changes nothing about what it showed.

The visible fight is over models, but a model is one layer in a stack that runs from energy to chips to infrastructure to applications. The real contest is architectural control across all of it, and the frontier labs are not defending a single layer. They are using their lead in models to orchestrate the whole stack, integrating down into infrastructure and up into deployment and services. Venkat Venkatraman's work on digital competition is the clearest account of why orchestration beats component quality, and I have taken that up at length in the argument that Europe's gap is capital and distribution rather than science. This piece is about the layer underneath that one: whether Europe can act as a single thing at all.

Two powers are contesting that control as single integrated markets. The United States does it with private capital at continental scale and one deep pool of risk money. China does it by the state, with one market and one direction. Each can orchestrate because each is, in the way that matters, one thing.

The third pole that cannot act as one

Europe is the only other mass large enough to be a third pole. It is also the only one that cannot act as one. The talent is here. The capital exists. The demand is here. What is missing is cohesion.

Mistral has been the standing proof of the first half of that. The science is not the problem: a European lab can build a frontier model, and has. What it cannot find at home is the capital depth to fund three more years of compute, or the integrated market to put that model in front of ten thousand enterprises at once. A subsidy fixes neither. It funds one more car and leaves the road unbuilt.

What cohesion means in practice

Europe is twenty-seven markets, with twenty-seven insolvency regimes, no common capital market, and household savings that sit in national banks instead of funding companies that could scale across the continent. You cannot orchestrate a stack across that. Mario Draghi's competitiveness report and Enrico Letta's single-market report, both from 2024, said the same thing in different words: the binding constraint is not ideas, it is fragmentation and the capital that fragmentation strands.

I spent years inside European regulated finance. The limit I watched was never the quality of the people. It was that a decision large enough to matter had to clear too many national lines to move at the speed the work demanded.

The asymmetry is the whole story. The United States never had to build cohesion. It was born integrated, one currency, one capital market, one bankruptcy code. China has cohesion by the state. Europe is the only bloc that would have to manufacture it on purpose, against twenty-seven national interests, and it has tried for a decade without finishing.

The objection: does Europe not have the data?

Here is the pushback I get, and it is a good one. Europe's incumbents, its banks and insurers and manufacturers, hold proprietary operational data the labs cannot scrape. Is that not a moat? It is real, and it is necessary. But it is not sufficient. Data is universal. Every incumbent everywhere holds some. The question is whether you can orchestrate around it, and orchestration needs scale and a single road. Without the engine, the same forward-deployment partners that carry the labs into your institutions absorb that data too, and Europe concedes the road a second time.

Third order, when the contest is fourth

This is where the layers beneath the headline matter. First order, a model does something useful. Second order, firms change how they work. Third order, institutions respond, with subsidies, with sovereignty panic, with export controls. Fourth order, the system reorganizes, or it fails to. Europe keeps acting at the third order, defensive and national, reaching for a champion, when the contest is fourth-order. The question is not which model you fund. It is whether you build the single market that can orchestrate the stack at all.

Cohesion has a name

The Capital Markets Union has been on the table since 2015 and has never been delivered. It is not a technical file. It is the measure of whether a fragmented mass can choose to become one. It is the one lever that turns Europe's talent and Europe's capital into a pole instead of a supplier.

The test you can run this quarter

Take the European AI positions you already hold, in your portfolio or in your vendor stack, and score each one on a single question. Not how good the technology is. Can this company reach ten thousand customers and raise three more rounds without leaving the continent?

Three answers come back, and each is a decision.

It can. Rare, and worth more than the model benchmark that got it on your list.

It cannot, and it is honest about needing a US capital round or a US hyperscaler to distribute. Then price it as an acquisition candidate, because that is what it is, and ask yourself what happens to your dependency when the acquirer is American.

It cannot, and the plan is a subsidy or a sovereignty mandate to close the gap. You are funding a car and calling it a road. That position needs a reason beyond the policy holding.

Run the same question on your own institution, because it is not only a vendor test. If your firm could not take a decision large enough to matter without clearing several national lines, you are inside the constraint, not observing it.

The position I hold

Europe does not have a model problem. It does not even have an engine problem. It has a cohesion problem, and cohesion is the one thing it would have to choose, deliberately, against its own fragmentation.

Which is why you cannot wait for that choice before acting on it. Allocate as though the fragmentation is permanent, and you will be right for longer than is comfortable. If Europe proves you wrong, nothing in that position is hard to unwind. That asymmetry is the whole of the advice.