For two years the consensus held that Europe had lost the AI race. Then the money started moving. Investor Nicole Junkermann on how to tell the companies that will build Europe’s AI future from the ones borrowing its language.
Europe’s favourite intellectual pastime is writing its own obituary. In artificial intelligence, the draft has been circulating for years: the models are American, the chips are Taiwanese, the capital is everywhere but here. It’s a tidy story. It’s also increasingly out of date.
Consider what has actually happened over the past eighteen months. Mistral, the Paris-based model developer, closed a €1.7 billion Series C, the largest European funding round of the year. Nearly a third of all venture capital invested in European startups in 2025 went to AI and machine learning companies. And in Brussels, the Commission launched InvestAI, an initiative to mobilise €200 billion for AI across the bloc, anchored by a €20 billion facility for up to five “gigafactories”, each housing more than 100,000 advanced AI processors. When the Commission asked who might want to build them, it received 77 proposals across 16 member states, with proposed investment north of €230 billion.
Even the regulation is changing shape. The AI Act – Exhibit A in every essay about European overreach – has just been formally recalibrated. Last month the EU gave final approval to its “digital omnibus” package, deferring the heaviest high-risk obligations to the end of 2027 and beyond, and lightening the compliance load for smaller companies. One can argue about whether the rethink came late. It’s harder to argue that Europe hasn’t noticed the problem.
The infrastructure turn
What’s emerging is a deliberate resequencing. Europe opened its AI decade with rules; it’s now racing to add the two things rules can’t summon on their own, which are compute and capital. Nineteen AI factories are being stood up across the continent on the back of Europe’s supercomputing network, with the gigafactories to follow from 2027, and the Commission’s stated ambition is to at least triple the EU’s data centre capacity within five to seven years.
Some scepticism is worth keeping to hand. Public compute programmes have a habit of arriving after the moment they were designed for, and the private capital gap remains uncomfortable: European AI companies raised around $14 billion last year against $146 billion in the United States, according to Atomico.
But governments rarely move at the speed of technology. They don’t have to. They only have to move quickly enough that the next generation of companies doesn’t leave before it starts. The point of subsidised compute isn’t to out-build California. It’s to stop the price of experimentation deciding, all by itself, where European ideas get built.
From models to problems
For investors, though, the more useful question isn’t whether Europe can fund a rival to the American labs. It’s where value will actually settle. The first phase of the boom rewarded model-builders; that trade is narrowing as frontier models become harder to tell apart and open alternatives improve almost monthly. Value is migrating to the infrastructure underneath, and to the applications above that own genuinely hard problems.
Nicole Junkermann, the founder of NJF Capital, has been investing in AI since well before the current cycle, and her filter is blunt. “The hardest thing in AI right now isn’t finding companies, it’s ignoring them,” she says. “Ninety per cent of what crosses my desk is a thin layer on somebody else’s model. The businesses that matter own hard problems – biology, logistics, energy – where the data is difficult and the moat is real.”
On this reading, Europe is better positioned than the obituaries allow, because hard problems are what the continent has in abundance. Its industrial base, its health systems, its energy transition: all of them generate the difficult, regulated, proprietary data from which application companies build moats. Junkermann’s own portfolio offers a worked example. NJF Capital was an early backer of Owkin, the Paris and New York-based company applying AI to drug discovery – a bet made when hospital data was widely considered too messy and too regulated to build a business on. Owkin has since grown a €90 million research partnership with Sanofi, extended last month into a multi-year agreement under which it will build AI agents for the French pharmaceutical group’s drug development work, backed by a five-year licence for K Pro, its “AI scientist” platform. AstraZeneca has signed a similar agreement. Regulated markets, difficult data, decade-long problems: what looked like Europe’s handicaps turn out to be the raw material of defensibility.
The honest ledger
None of this settles the harder accounts. Europe generates 17% of new global enterprise value in technology but captures just 10% of global exit value, which is another way of saying it grows companies for other people’s benefit. Adoption lags too: only 13.5% of EU companies use AI at all. And the talent trained here still drifts west with depressing regularity.
Nobody should mistake eighteen good months for a settled outcome, and the obituary may yet be republished; Europe has form. But for the first time in the short history of this technology, the continent’s investors have something better to do than mourn – and rather a lot to choose from.
