Europe’s frontier AI gap is now a power-and-capital problem

The gap has changed shape

Europe’s frontier AI problem is often described as a talent problem. The latest advice to the European Commission lands somewhere else.

The continent has good researchers. It has universities, industrial expertise and a large market. What it does not yet have, at the required scale, is enough computing capacity, enough dependable energy for that capacity, or enough late-stage money to keep ambitious AI companies growing in Europe.

That is the central message from a new report published by the European AI Office on Wednesday. It brings together findings from more than 100 experts working across AI development, industry, investment, research, government and civil society.

The report is not EU policy. It is a set of recommendations, and the Commission is careful to say that the conclusions do not represent its official position. Still, it gives a fairly blunt account of where Europe is falling behind — and where the next two years may matter most.

Europe is not starting from zero. The experts describe a region with world-class research, a sizeable pool of technical talent and strengths in several parts of the technology stack. It also has a deep industrial base, from manufacturing to healthcare and energy.

But most frontier model development still happens elsewhere, largely in the United States and China. Turning European research into companies that can train and operate leading systems has proved much harder.

The report’s point is simple: scientific ability alone does not buy thousands of advanced chips, connect a data centre to the grid or carry a young company through an expensive growth phase.

That changes the debate. Europe’s shortfall is not only about inventing better models. It is about the physical and financial machinery around them.

Compute needs power

The experts put compute infrastructure and energy at the top of the two-year list.

Those two things are tied together. New AI facilities need large quantities of electricity, but they also need grid connections, cooling, land and permits. A plan for more computing capacity means little if the power cannot arrive on time or at a workable price.

The EU has already made AI Factories and proposed AI Gigafactories part of its broader strategy. The new report adds urgency. In the experts’ view, Europe has a narrow window to make the infrastructure real, not simply announce it.

There are questions the report cannot answer on its own. Where will the facilities be built? How much public money should support them? Who gets access? And how will new demand sit alongside Europe’s energy and climate goals?

Those choices will shape whether public infrastructure widens access to serious computing power or mainly helps a small number of already well-funded organisations.

The missing middle of finance

Money is the other large constraint.

Europe can fund research and early-stage companies. The harder stage comes later, when a company needs hundreds of millions — sometimes more — to train models, hire internationally and sell across several markets.

The experts call for more growth-stage capital. Without it, a promising European company may move its centre of gravity abroad, sell early or depend on infrastructure controlled elsewhere.

That does not mean every AI company should be kept European at any price. It does mean that Europe cannot claim technological sovereignty while treating company scale-up as someone else’s job.

Rules still matter, but clarity matters too

The report also points to legal uncertainty around training data, copyright and data protection.

This is not a request to remove every rule. It is a request for companies and researchers to know what the rules mean in practice. Long periods of uncertainty can be manageable for large incumbents with legal teams. They are harder for smaller organisations deciding where to build.

Europe’s regulatory reach is itself one of the assets identified by the experts. A large market with trusted rules can be valuable. But that advantage weakens when compliance is difficult to interpret or changes slowly arrive in usable guidance.

Sovereignty is not isolation

The report recommends looking across the whole AI stack — chips, cloud, data, models, applications and energy — to find where Europe has real leverage.

That is more realistic than trying to reproduce every part of the global supply chain inside the EU. Europe will still need partners. The question is where dependence is acceptable and where it creates a strategic risk.

The experts point to the size of the European market, its industrial base, its regulatory influence and its position in parts of the technology stack. Those are useful cards. They are not a frontier AI ecosystem by themselves.

What happens next

The AI Office says it will prioritise the recommendations and work towards implementation. That next step matters more than the report.

The useful tests are quite concrete: whether computing capacity comes online, whether enough clean and reliable power is available, whether European companies can raise large rounds without leaving, and whether legal guidance becomes easier to use.

The report does not settle any of those issues. It does narrow the argument. Europe knows how to produce AI research. The question now is whether it can build the conditions that let that research grow at home.