The Emigration Question: Is European AI Actually Leaving?

Every new European regulation generates a wave of “brain drain” predictions. GDPR was supposed to drive data-intensive startups to the US. The Digital Services Act was supposed to make social media companies abandon the European market. Each time, the predictions are approximately half right: some companies leave, most adapt, and the net effect is more complicated than the dire warnings suggested.

The EU AI Act has generated its own brain drain discourse, particularly since enforcement began in earnest in early 2026. The narrative runs roughly as follows: European AI talent and startups, facing compliance overhead that American and Asian competitors don’t, will migrate to more permissive jurisdictions. San Francisco will see an influx of European AI founders. London (post-Brexit, with UK’s lighter-touch AI approach) will benefit. Dubai, Singapore, and Austin will compete for the overflow.

The narrative has evidence in its favor. It also has significant evidence against it. And the more important question — not who is leaving, but who is choosing not to come — gets far less attention.

What the Movement Data Actually Shows

Tracking founder and researcher movement is methodologically difficult. People don’t file emigration papers when they move. Startups don’t announce they’re relocating specifically due to regulation. LinkedIn data and visa statistics are the best proxies available.

UK Tech Nation’s June 2026 AI Talent report found that UK AI sector employment grew by 31 percent year-over-year, with approximately 18 percent of new hires coming from EU member states. This represents acceleration relative to the 12 percent EU-origin share in 2023-2024. The UK is attracting more European AI talent post-Brexit, though the UK’s own AI regulatory framework and cost of living are both relevant factors alongside EU AI Act compliance concerns.

Related reading

US visa data shows an increase in O-1 visas (extraordinary ability) granted to European AI researchers and founders — 847 in fiscal year 2025, up from 634 in fiscal year 2024. This is real but not dramatic. The NVCA (National Venture Capital Association) found in its July 2026 survey that approximately 4 percent of US AI startup founders who moved from Europe in the last 18 months cited EU AI Act compliance as a primary reason for relocation. Regulatory burden of all kinds was cited by 23 percent as contributing factor.

These numbers support the proposition that some movement is happening and that regulation is part of it. They don’t support a dramatic exodus. The more telling data points concern not movement but founding decisions.

The European Innovation Council’s July 2026 report found that the number of new AI startups incorporated in EU member states in H1 2026 was down 14 percent from H1 2025. This decline is not uniformly distributed: it’s most pronounced in high-risk application verticals (healthcare, financial services, employment) and least pronounced in minimal-risk verticals (creative tools, gaming, productivity). People are still founding AI companies in Europe. They’re founding fewer companies in the areas where AI Act compliance overhead is most significant.

The Non-Arrival Problem

The brain drain narrative focuses on who leaves. The more structurally significant question is who doesn’t arrive.

European AI ecosystem strength depends significantly on global talent attraction — researchers and engineers who could choose to build anywhere choosing to build in Berlin, Paris, Amsterdam, or Warsaw. In a competitive global talent market for AI capability, a regulatory environment that increases compliance overhead and creates legal uncertainty is a negative factor in the location decision.

A survey of 200 senior AI researchers at US universities, conducted by the European Research Council in May 2026, found that 61 percent believed the EU AI Act had made European research positions “somewhat” or “significantly” less attractive relative to two years ago. This doesn’t mean they won’t come. It means the EU needs to offer more compelling reasons to come — funding, research infrastructure, visa facilitation, compensation — to offset the regulatory negative.

The same dynamic operates at the company formation level. A founder choosing between US incorporation and EU incorporation is making a bundle of decisions about market access, investor availability, regulatory environment, talent access, and operational infrastructure. The EU AI Act has changed one of these factors at the margin. Whether it changes the overall choice depends on the relative weight of all factors.

For founders targeting the European enterprise market specifically — regulated industries with EU customers who have their own compliance requirements — the EU regulatory environment is less disadvantageous, because the compliance cost is either unavoidable (their customers require it of vendors) or is a moat against non-EU competitors who haven’t built the compliance capability.

For founders building AI applications with global market potential who don’t specifically need EU-first deployment — the US incorporation option has become marginally more attractive. This is a real shift at the margin, not a transformation.

The Deeper Structural Problem

The emigration question, while contested, obscures a deeper structural concern that has less to do with the AI Act specifically and more to do with what it represents.

Related reading

The EU AI Act signals a European regulatory philosophy that treats AI development primarily as a governance and rights problem rather than an industrial competitiveness problem. This is a coherent philosophical position — the EU’s approach to digital regulation generally prioritizes citizen rights and democratic governance over economic competitiveness in a way that reflects genuine political values. GDPR, DSA, DMA, and now the AI Act all express this prioritization.

The problem is that AI is not primarily a governance problem right now. It is primarily a competitiveness problem. The countries that build the best AI at the most useful scale over the next decade will have structural advantages in productivity, military capability, scientific research, and economic output that compound over time. The US and China understand this in a visceral way that shapes their AI policy. The EU understands it intellectually but is organized, institutionally, around the governance prioritization.

This isn’t an argument that the AI Act is wrong. The rights and governance concerns it addresses are legitimate and important. It is an argument that the AI Act, however well-designed, operates within a broader European regulatory philosophy that may produce competitiveness consequences independent of any specific provision’s design.

The GDPR analogy is relevant here, but in a different way than the usual comparison. GDPR did not drive a dramatic European exodus of data companies. It did contribute, alongside a range of other factors, to a European digital economy that is structurally weaker relative to the US than it was in 2015. European consumers’ data is better protected. European data companies have smaller market positions relative to US counterparts. Both things are true simultaneously.

The EU AI Act may produce a similar outcome: European citizens with better protections against harmful AI, European AI companies with smaller market positions relative to US and Chinese counterparts. Whether this is the right trade-off is a genuinely difficult political question that the regulation’s proponents have not engaged with as directly as its opponents deserve.

Who Stays and Why

The people not leaving — the majority of European AI talent and founders who are adapting to the Act rather than emigrating — deserve as much attention as the ones who are.

The European AI ecosystem has genuine strengths that the compliance overhead doesn’t negate: world-class research universities, strong engineering talent pipelines, established enterprise customer bases, and for certain verticals (industrial AI, healthcare information systems, enterprise software) a deep understanding of regulated market dynamics that US competitors struggle to replicate. These advantages don’t disappear because conformity assessments are expensive.

Some European AI founders are explicitly choosing to build in Europe because they believe the regulatory environment, however demanding, will eventually produce a more trustworthy and sustainable AI industry. This is a bet on the proposition that the first mover in compliant, trustworthy AI has a long-term advantage as global AI governance converges. The bet may be wrong. It may be right. The people making it are not fleeing. They’re investing.

The emigration question has a definitive answer that nobody honest wants to give: some talent and startups are leaving, more than would have left otherwise, but not enough to constitute an exodus, and the more consequential problem is the slower-to-observe question of who doesn’t arrive. The EU can probably maintain a competitive AI ecosystem. It cannot take for granted that the ecosystem will reach the scale and ambition that European industrial policy needs without addressing the regulatory overhead that makes it marginally harder to build significant AI here than anywhere else.

The AI Act is not a fatal blow. It is one more weight on a scale that was already not tipping in Europe’s favor.

Related reading

The Policy Response Gap

European institutions are not ignoring the competitiveness concerns. The European Commission’s June 2026 “AI Competitiveness Package” included a €2 billion commitment to AI supercomputing access for European researchers and startups, expanded EIC (European Innovation Council) fast-track funding for AI scale-ups, and new visa facilitation measures for non-EU AI talent relocating to Europe.

These are meaningful investments. They also reveal the nature of the problem: the policy response to competitiveness concerns about the AI Act is not to modify the AI Act but to compensate for it with subsidies and talent programs. This is not irrational — changing a newly enacted regulation creates uncertainty and damages regulatory credibility — but it means the compliance overhead remains while the compensation mechanisms are added on top.

The €2 billion in supercomputing access addresses a real gap but isn’t primarily relevant to the compliance overhead problem facing health, fintech, and employment AI startups. Compute access helps with training large models. It doesn’t help a Vienna healthtech company pay its Notified Body bills or a Dutch HR tech startup navigate product reclassification. The policy tools being deployed address different problems than the ones the emigration debate is actually about.

There’s a version of this story that has a good ending — where the compliance overhead normalizes, the Notified Body capacity expands, EU-compliant AI becomes a genuine global competitive advantage as other jurisdictions converge toward similar standards, and the founders who built compliant AI in Europe capture the value as global enterprise customers start requiring it. This scenario is plausible and some people who have thought carefully about it find it probable.

The honest version acknowledges that it requires things to go right that could go wrong: fast enough Notified Body expansion, sufficient funding for early-stage AI in high-risk verticals, successful CJEU harmonization of enforcement divergence, and continued European enterprise demand for AI that can be purchased with regulatory confidence. None of these are guaranteed. The emigration question will be answered, definitively, by which scenario actually plays out over the next five years. The early evidence suggests a muddled middle — neither catastrophe nor vindication — which is probably what both sides in this debate least want to hear.