There is a particular kind of dread that has settled over European policy circles in recent years, and it is not the sharp panic of a sudden crisis but something closer to the slow-building anxiety of watching a race and realizing you started it on the wrong track. The question being asked in think tanks in Brussels, in parliamentary committee rooms in Strasbourg, in tech conferences from Paris to Berlin, is more or less the same: is Europe being left behind in AI, and if so, what can anyone actually do about it?
The unease has a very specific shape. It is not about whether AI is dangerous — Europeans have strong opinions on that, and plenty of legislation to prove it. It is about whether, in the process of worrying so much about AI’s risks, Europe has managed to sit out the era of AI’s creation. Mario Draghi, whose 2024 competitiveness report has become the closest thing Brussels has to a shared scripture, put it plainly: the productivity gap between the EU and the United States is largely explained by the tech sector, or by Europe’s relative lack of one. There is no EU company with a market capitalisation above €100 billion that has been built from scratch in the last fifty years. Every single one of the six American companies now worth over a trillion euros was founded in that same period.
That is not a temporary gap. That is a different history.
The cloud market, everyone quietly agrees, is already gone. Around 70% of foundational AI models since 2017 have been developed in the United States, and just three American hyperscalers control over 65% of the European cloud market. So when a startup in Amsterdam or a research lab in Munich needs to train a model, it is almost certainly doing so on someone else’s infrastructure, paying in dollars, and feeding data into systems governed by US law and US corporate priorities. Before a single line of proprietary code creates value, a European AI company is already a tenant in someone else’s building — and the landlord is increasingly also its competitor.Fortune
Then there is the talent problem, which is perhaps even harder to fix because it is not really a policy failure so much as a gravitational force. Europe’s AI workforce is exceptionally well-educated and internationally mobile — on average 57% of AI professionals completed their undergraduate studies outside Europe, compared to 38% in the US. Those internationally mobile professionals tend to move again, and the destination is usually America. Net tech talent inflows to Europe fell by half between 2022 and 2024, from around 52,000 to just 26,000. By 2025, 62% of EU AI postdocs said they were planning to leave for the US or China, drawn by salaries and projects that Europe’s fragmented, underfunded ecosystem simply cannot match. The continent trains people brilliantly, then watches them board planes to San Francisco.
What makes all of this particularly awkward is that Europe’s response to the AI moment was, for years, primarily regulatory. The EU AI Act was presented not as a concession to others’ technological leadership but as a form of soft power — the argument that Europe could shape how the world governs AI even if it couldn’t dominate how AI was built. Brussels had done this before with data protection, and GDPR had indeed become a global reference point. Why not do it again?
The problem is that governing a technology and building it are not the same thing, and the companies doing the building noticed the difference. More than two-thirds of European companies surveyed by Amazon Web Services said they struggled to understand their responsibilities under the AI Act. Over 45 European companies, including Airbus and Mistral AI, signed letters calling for postponement of the law. The Swedish prime minister called the rules “confusing.” The implementation guidance for prohibited AI systems arrived two days after those prohibitions took effect, leaving companies with essentially no time to prepare. Mexc
The capitulation, when it came, was swift and striking. In July 2025, a Commission spokesperson had said clearly: “There is no stop the clock. There is no grace period. There is no pause.” Four months later, EU officials were openly negotiating exactly that. The deal reached in May pushed back enforcement in high-risk AI sectors — biometrics, critical infrastructure, employment, health — by sixteen months, to December 2027. AI embedded in physical products like lifts and toys gets an even longer runway, until August 2028. Officials insisted this was about sequencing, not retreat. The rest of the world interpreted it as retreat.
What has replaced the regulatory-first posture is something more explicitly industrial. The AI Continent Action Plan calls for up to five AI gigafactories — large-scale compute facilities designed to train frontier models at a scale Europe currently cannot attempt — backed by a €20 billion investment vehicle called InvestAI. At Davos in 2026, Emmanuel Macron declared that Europe needs “more sovereignty and more autonomy,” and that investment must accelerate dramatically in AI, quantum, and defence. The language of sovereignty has become ubiquitous in European AI policy, a word that acknowledges dependency while promising escape from it.
Whether the escape is actually possible is a harder question. European industry pays roughly double the electricity rates of American counterparts, and with large model training runs costing hundreds of millions of euros, that structural disadvantage compounds at every scale. Five flagship infrastructure projects now account for over €30 billion in targeted compute investment — a concentration that creates serious delivery risk, since any failure to execute on time leaves the entire strategy without a fallback. European AI funding did reach a record $21.8 billion in 2025, up 58% in a single year, which is genuinely significant — but the starting point matters, and the walled gardens of American platform infrastructure are not going to dissolve because European venture capital has had a good year. Deep Tech
The United States produced 40 notable AI models in 2024. Europe produced three. That ratio is not primarily a consequence of regulation, or even of funding. It reflects decades of divergence in how universities, capital markets, and technology ecosystems are structured — divergence that no action plan, however ambitious, can close in a legislative cycle.
None of this means Europe is finished. Where trust, interoperability, and domain-specific depth define long-term value — in industrial AI, in healthcare, in autonomous systems — European companies still have genuine ground to compete on. And there is a real, if unexpected, opportunity opening up as American visa policy tightens and some international AI talent begins looking elsewhere. The EU and India Free Trade Agreement, signed in early 2026 with a companion mobility framework, is expected to accelerate the flow of Indian-trained AI talent into Europe — potentially changing the composition of the continent’s technical workforce in ways that matter. HCSS
But what Europe cannot afford is the comfort of believing that getting the governance right is the same thing as winning. The anxiety in Brussels is real and, on the evidence, justified. The question now is whether it produces something more than better-organized plans for things that take a decade to build, in a race where the leaders are already several years ahead and accelerating.
Subscribe to EuroLuminant for independent European journalism.
