The United States has produced 40 AI foundation models. China has developed 15. All of Europe combined has created just three.
The European Union is losing the global “AI race” on nearly every key metric except regulation. While China and the United States invest billions in infrastructure, talent, startups, labs, and research, Europe remains focused on rules. Policy burdens and fragmentation across 27 member states create major hurdles: progress is inconsistent, talent leaves, and capital goes elsewhere.
Clark Parsons, leader of the European Startup Network, is blunt about the imbalance. “The EU should stop patting itself on the back for being the world’s regulator in technology. Some elements of the Digital Markets Act were designed to promote competition. I like those, but in general we have spent far too long focusing on regulating instead of every day waking up and saying what can we do to make Europe the most competitive place on the planet, the most prosperous place on the planet.”
“If I had to say, ‘Please stop doing one thing,’ I would say, ‘Stop thinking about how to regulate and start thinking about how to unleash incredible growth,’” he adds.
Parsons also questions whether regulation is the best way to guarantee trust in a fast-moving technological field. “The AI world is moving so fast. It’s hard to see what’s coming. I think clever entrepreneurs and technologists are going to deliver ways for us to establish trust and establish safeguards.”
Despite its current position, the EU refuses to concede defeat. As part of its 2025 AI strategy, European Commission president Ursula von der Leyen promised that “from now on, it’s ‘AI first’”, vowing to “spare no effort to make Europe an AI continent”.
“The AI race is far from over. We are only at the beginning, and global leadership is still up for grabs,” she declared at the Paris AI Action Summit in February 2025.
Talent without traction
The paradox is clear. Although Europe produces top talent, it fails to retain it. The EU has about 30% more AI professionals per capita than the US, but better funding, clearer career paths, and softer regulations abroad lure them away. Three out of four European international AI PhD students at American universities stay in the US for at least five years. In total, a third of non-US AI specialists move to the United States.
This talent drain raises a fundamental question: has Europe already lost the global race on AI?
“When it comes to AI startups and scale-ups in Europe, there’s very clearly some hurdles. And if I had to boil it down to one, I would say it’s finance and financing,” Parsons says.
The United States invests four to ten times more in AI than the EU. Annual AI venture investment in the US is $60–70 billion, compared to about $7–8 billion in the EU. Over the past decade, private AI investment in the US exceeded $400 billion, while all EU countries combined attracted about $50 billion.
According to Parsons, “[the US] also has extremely deep pools of capital. You see how comparatively easy it has been for OpenAI to raise enormous sums. Other new entrants, like Anthropic, got incredible valuations and incredible amounts of capital.”
Infrastructure gaps and late catch-up
This funding gap directly affects Europe’s AI infrastructure. The continent has fewer data centres and much less AI-specific compute capacity. To address this, the European Commission has announced initiatives, including AI “factories” and future “gigafactories” with many accelerators, backed by public funding and expected private co-investment.
Through its InvestAI initiative, the EU aims to mobilise €200 billion, including €20 billion for the construction of up to five AI gigafactories, each expected to produce more than 100,000 advanced AI chips. EuroHPC has already received 76 proposals from 16 countries to host these facilities, and Brussels aims to triple Europe’s data-centre capacity within five to seven years.
Beyond infrastructure, the EU has steadily increased funding for AI. Through Horizon Europe and Digital Europe, the Commission already allocates more than €1 billion each to AI. The AI Continent Action Plan mobilised €20 billion for AI scaling in April 2025, followed by €1 billion under the Apply AI Strategy in October 2025.
These European projects are still under construction, while US cloud providers already operate hyperscale clusters for AI workloads. Even Europe’s most powerful supercomputers are better suited to traditional high-performance computing than to large-scale AI training, after years of underinvestment in AI-specific infrastructure.
Venture capital and startup exodus
European venture capital is structurally more cautious than in the US. AI startups in Europe raise about $8.5 million in their first funding rounds, compared with $13 million in the US. US venture capital firms manage roughly $270 billion, six times more than the $44 billion managed in Europe.
These differences make it harder for European startups to grow, adopt AI at scale, and retain talent. They also influence where companies choose to base themselves.
Parsons points to a telling example. “Let’s look at Lovable, the fastest-growing AI company in Europe, based in Stockholm. The founder is Swedish. His team is Swedish. The angel investors are Swedish. But the company is legally registered in Delaware. And this is just because the access to capital is so much easier in the US.”
Mobility within Europe is also limited. “Only about 18% of our venture capital crosses borders now in Europe,” Parsons explains. “So, if you’re sitting in Paris or Munich or London or Stockholm, you’ve got a pretty nice pool of local investment money. But if you’re sitting in Barcelona or Lisbon or Milan, or Bucharest, it’s harder… and you might have to leave or move.”
Regulation, fragmentation, and the AI Act
Regulation remains a central challenge. Europe wants to be a global leader in ethical, human-centric AI. By August 2027, the European Commission plans to implement what it calls the world’s first comprehensive AI regulation.
At the heart of this effort is the AI Act, which is based on a risk-based approach: the greater an AI system’s potential impact on people, the stricter the rules governing it. The Act sets requirements for AI providers and deployers to prevent harms such as manipulation, discrimination, intrusive biometric profiling, deepfakes, and social scoring, with the stated aim of ensuring trust in AI systems.
Enforcement is inconsistent and insufficient. While some member states like Italy, Spain, Denmark, and Ireland are making significant headway in the AI Act’s application, others still lack fully operational enforcement bodies, putting the immediate impact of the AI Act at risk and missing Brussels’ intentions.
Critics argue that the EU’s strict rules and bureaucratic complexity have slowed innovation. International companies have also asked the Commission to ease aspects of the framework. As the AI Act introduces legal uncertainty, its scope must be “proportionate and support innovation and development,” warned economist Mario Draghi.
For startups, the effects are tangible. European AI companies face enterprise sales cycles that are 30% longer than in the US, deal sizes that are 50% smaller, and higher expansion costs, largely due to regulatory fragmentation across 27 national markets. Unlike the US or China, the EU lacks a single, unified market for AI deployment.
Fragmentation also affects data. Differences in privacy enforcement, sector-specific rules, and public-sector data-sharing practices make it difficult to build continent-wide datasets. Developers in some member states say varying interpretations of GDPR and copyright law limit which datasets they can use. As a result, companies often rely on non-EU data or foreign AI models trained elsewhere.
The trend is unmistakable. Swedish AI companies, such as Sana Labs, end up acquired by US firms. Stockholm produces many unicorns per capita, but founders consistently turn to American investors for scaling.
“It is hard right now to scale across Europe. We have very different markets, with no single market for startups or scale-ups. If you start here, you generally have a tougher time than in one giant market like China or the United States,” Parsons says.
Dependency on the US and China
For now, Europe depends heavily on external players for the core components of AI. The world’s leading large language models are American or Chinese. European companies rely on platforms they do not control.
US hyperscalers dominate cloud and compute in Europe. Amazon Web Services (32%), Microsoft Azure (23%), and Google Cloud (10%) together hold 65% of the European cloud market. Overall, US providers control around 72%, while EU-based companies account for less than 20%. The US has 17 times Europe’s AI supercomputing capacity and controls 74% of global high-end AI compute.
Most advanced AI chips are designed and made outside Europe, mainly in the US and East Asia. China leads in AI patents and is advancing quickly in generative AI, shaping global standards and competition.
A race still open, but narrowing
Facing criticism, the European Commission has begun to signal a shift. In November last year, it launched a review of the rules governing digital innovation, the Omnibus revision of the Digital Rulebook. The aim is to simplify parts of the AI Act and related legislation to boost competitiveness and accelerate AI development.
While the European Parliament and Council continue to discuss, the Commission has already proposed more simplification. It is not yet clear if this will lead to faster scaling and more investment. The race is not over, but the EU’s window to catch up is closing quickly.
