Redistributed

Europe has the world's most comprehensive AI regulation. It does not have a single frontier AI lab.

This is not an oversight. It is a structural condition with deep redistribution consequences. The European Union writes the rules for AI systems built in San Francisco and Beijing, deployed on American cloud infrastructure, funded by American and Asian capital. It is the world's most powerful consumer of AI and the world's most sophisticated regulator of AI. It is not, in any meaningful sense, a producer.

The question this raises is not the one usually debated — whether regulation stifles innovation. It is more fundamental: when you regulate technology you don't build, who captures the value? And when your "national champions" depend on foreign capital and foreign compute, what exactly is sovereign about them?

The Gap

Start with the numbers. The five companies with the largest AI research budgets — Alphabet, Microsoft, Meta, Amazon, and Apple — are American. The most valuable Chinese AI companies — ByteDance, Baidu, Tencent — would form a second tier. Europe's entrants do not appear on either list.

Mistral AI, founded in Paris in 2023 by alumni of Meta and Google DeepMind, is the continent's most prominent AI company, valued at roughly $6 billion. AMI Labs, also in Paris, raised a $1.03 billion seed round in early 2026 — one of the largest in AI history. The UK has DeepMind, arguably the world's most accomplished AI research lab. But DeepMind has been owned by Google since 2014. Its profits flow to Alphabet shareholders in Mountain View.

What Europe has is considerable: world-class research institutions (INRIA, Max Planck, Cambridge, ETH Zurich, EPFL), a deep talent pipeline, multilingual capability, strong privacy frameworks, and public funding mechanisms like Horizon Europe. What it lacks is equally clear: hyperscale compute, venture capital at American scale, and platform companies that generate the data flywheels frontier AI development requires.

This is the paradox. Europe produces researchers who advance the field, then watches them leave — or watches the companies they build get acquired by American buyers. As the brain drain research shows, public universities across Europe train talent at taxpayer expense, then lose them to labs offering salaries that no European institution can match (The Academic Brain Drain: Who Pays to Train AI's Best Minds?).

The French AI Moment

If Europe has an AI capital, it is Paris. This is not accidental. It is the product of deliberate industrial policy.

President Macron has invested heavily in positioning France as an AI leader. The Jean Zay supercomputer, operated by CNRS, provides public compute for French researchers. Station F, the world's largest startup campus, hosts AI ventures. The France 2030 plan committed billions to emerging technologies, with AI as a centrepiece. The message has been consistent: France will build, not just regulate.

The results are real. Mistral emerged from this ecosystem — its founders, Arthur Mensch, Guillaume Lample, and Timothée Lacroix, came from Meta's FAIR Paris and DeepMind. AMI Labs, led by Yann LeCun — a foundational figure in deep learning who left his role as Meta's chief AI scientist — chose Paris as its headquarters. Xavier Niel, the telecoms billionaire behind Free and Iliad, has become a one-man AI industrial policy, backing multiple ventures and providing infrastructure.

But look closer at the capital structure and the sovereignty claim gets complicated. Mistral took a $16 million investment from Microsoft in early 2024, part of a broader partnership that included distributing Mistral's models through Azure. AMI Labs' $1.03 billion seed round was led by Jeff Bezos, Eric Schmidt, Nvidia, and Temasek — American and Singaporean capital. The company is headquartered in Paris. Its funding comes from everywhere but France.

Mistral's lobbying during the EU AI Act negotiations further complicates the picture. As documented in analysis of AI industry influence, Mistral pushed for lighter regulation of foundation models — aligning its interests with OpenAI and Google rather than with the European Parliament's original, stronger proposals (The Influence Machine). France backed Mistral's position, pressuring the Council to weaken foundation model provisions. The national champion, it turned out, wanted the same rules as the American incumbents.

This is not unique to France. It is the logic of competition: a French AI company competing with American ones needs the same regulatory environment. But it exposes the tension at the heart of European AI strategy. You cannot simultaneously champion a national AI industry and champion stringent AI regulation, because your national champion will lobby against the regulation.

The EU AI Act: Power Without Production

The EU AI Act, which entered force in August 2024, is genuinely significant. It is the world's first comprehensive, legally binding AI regulation. It creates a risk-based classification system, bans certain AI practices outright, and imposes substantial obligations on high-risk systems (The EU AI Act Explained).

As Anu Bradford has documented, the EU's regulatory influence extends well beyond its borders through the "Brussels Effect": companies find it easier to adopt EU standards globally than to maintain separate systems for European and non-European markets (Bradford, 2020). GDPR reshaped privacy practices worldwide. The AI Act may do the same for AI governance.

This is real power. It is the power to set terms. European citizens gain protections that American citizens do not have: the right to know when they are interacting with an AI system, restrictions on biometric surveillance, requirements for human oversight in high-risk applications. Workers gain new rights regarding AI systems used in employment decisions. These are substantive redistribution victories — transfers of power from companies to affected individuals.

But regulation without production creates a specific redistribution pattern. The compliance costs of the AI Act — legal teams, documentation, conformity assessments, auditing — are real. They fall on companies operating in Europe. The companies best equipped to absorb these costs are the largest ones: Google, Microsoft, Meta, OpenAI. Smaller European competitors face the same requirements with a fraction of the resources.

Daron Acemoglu and Simon Johnson have argued that the direction of technological change is a political choice, not a market inevitability (Acemoglu & Johnson, 2023). If Europe's primary AI policy lever is regulation, it can shape how AI is used within its borders — but it cannot shape what gets built. It can require transparency in AI hiring systems deployed in Germany, but it cannot ensure that a German company builds the system rather than licensing it from an American provider.

The result is a specific form of dependency. Europe pays compliance costs in Brussels and licensing fees in San Francisco. It captures the overhead of regulation without the returns of innovation.

Sovereignty or Real Estate?

The sovereign AI movement has given European governments a compelling narrative: build domestic AI capability to reduce dependence on foreign providers (The Sovereign AI Movement). France, Germany, and the UK have each invested in national compute infrastructure. The EuroHPC Joint Undertaking pools resources across member states.

But sovereignty requires more than a headquarters address. When AMI Labs is "headquartered in Paris" but funded by Bezos and Schmidt, built on Nvidia GPUs, and likely trained on American or Asian cloud infrastructure, the sovereignty is primarily jurisdictional. The company pays French taxes and employs French workers. These are not nothing. But the controlling economic relationships — capital, compute, cloud — run through Silicon Valley.

This pattern has a historical analogue. During the Cold War, European nations hosted American military bases. The bases provided security and local employment. They did not provide sovereignty. The decisions about deployment, strategy, and escalation were made in Washington. European AI "sovereignty" risks the same structure: local benefits, remote control.

Kate Crawford's observation applies: AI infrastructure is extractive, requiring land, energy, and labour, and the costs distribute differently from the benefits (Crawford, 2021). A data centre in Marseille extracts French electricity and occupies French land. The model it trains may generate returns primarily for investors in Palo Alto.

The deeper problem is compute. Training frontier AI models requires thousands of high-end GPUs, manufactured almost exclusively by Nvidia (an American company) and fabricated by TSMC (Taiwanese). Europe has no equivalent. It has no hyperscale cloud provider. AWS, Azure, and Google Cloud dominate European cloud infrastructure. When European AI companies scale up training runs, they rent American compute (AI Chip Export Controls).

What Europe Could Do Differently

Europe's assets are real, even if they are currently mismatched to the frontier AI race. The question is whether they can be converted into something more than raw material for American companies.

Public AI infrastructure. Mariana Mazzucato's framework for the entrepreneurial state applies directly: if the public funds the research and trains the talent, the public should capture some of the returns (Mazzucato, 2021). Europe could invest in public compute infrastructure — not just research supercomputers, but production-grade AI systems available to European companies and institutions at cost. This would reduce dependency on American cloud providers and create a genuine alternative to the private compute oligopoly (Public AI).

Leveraging regulation as industrial policy. The AI Act could be more than consumer protection. If Europe required that high-risk AI systems deployed on European soil meet interoperability standards, use auditable architectures, or store training data within European jurisdiction, it would create structural advantages for European providers. This is what China does, unapologetically. Europe does it tentatively, constrained by free-market ideology and WTO commitments.

Retaining talent. European universities produce excellent AI researchers. Retaining them requires competitive compensation, adequate compute, and meaningful problems to work on. France's public compute investments through CNRS are a start. Germany's investment in AI research centres addresses part of the problem. But as long as a postdoc in Munich earns a quarter of what Google Brain offers in London — which itself pays less than the Bay Area — the talent pipeline will leak (The Academic Brain Drain: Who Pays to Train AI's Best Minds?).

Building on genuine advantages. Europe's multilingual reality is an asset in a world where AI systems need to work across languages. Its privacy framework, while sometimes treated as a burden, creates trust that has commercial value. Its tradition of public service delivery — healthcare, education, transport — creates use cases where AI could deliver public benefit rather than private profit. These advantages will not produce the next GPT. They could produce something more valuable: AI systems that work for citizens rather than shareholders.

The Redistribution Ledger

Europe's AI position creates a distinctive redistribution pattern.

What flows in: Regulatory power. Citizens gain rights against AI systems through the AI Act. Workers gain protections. The "Brussels Effect" extends these protections globally. This is genuine and significant.

What flows out: Talent, trained at public expense, migrating to American labs. Licensing fees, paid to American AI providers. Cloud computing revenue, flowing to AWS, Azure, and GCP. Data, generated by European users, used to train American models.

What stays ambiguous: Whether European AI labs like Mistral and AMI Labs represent genuine sovereignty or the local franchise of a global industry. Whether regulation creates a protected market for European alternatives or just adds costs that entrench American incumbents. Whether Europe's research excellence eventually translates into industrial capability or remains an export product.

The paradox is real but not permanent. Europe regulated telecommunications and built Ericsson and Nokia. It regulated aviation and built Airbus. It regulated pharmaceuticals and built a globally competitive industry. In each case, regulation and industrial policy worked together — the rules created the market conditions, and sustained public investment created the companies. These are not just historical footnotes — they are proof that Europe's model can produce globally competitive technology industries when political will sustains the investment.

The question is whether European politics can sustain AI investment at the required scale through electoral cycles, fiscal constraints, and competing priorities. Frontier AI development is extraordinarily expensive, and the returns are uncertain. American venture capital tolerates this uncertainty because the potential returns are enormous. European public funding operates under different incentive structures — ones that have historically favoured caution.

Europe is not out of the AI race. But it is running a different race than the one it claims to be in. Its strength is governance, not production. Its leverage is market access, not technological capability. Whether it converts this leverage into genuine AI sovereignty — the kind that serves citizens, not just hosts companies — depends on choices being made right now, in Brussels, Paris, Berlin, and London.

The alternative is clear enough. Europe becomes the continent that writes the rules, pays the compliance costs, trains the researchers, and watches the value flow to San Francisco. It would not be the first time a regulatory superpower discovered that the power to regulate is not the same as the power to build.

Sources

Acemoglu, Daron, and Simon Johnson. Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity. PublicAffairs, 2023. Publisher
Bradford, Anu. The Brussels Effect: How the European Union Rules the World. Oxford University Press, 2020. Publisher
Crawford, Kate. Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press, 2021. Publisher
European Parliament and Council. Regulation (EU) 2024/1689 (the AI Act). Official Journal of the European Union, August 2024. EUR-Lex
Mazzucato, Mariana. Mission Economy: A Moonshot Guide to Changing Capitalism. Penguin, 2021. Publisher
OECD. "National AI Policies and Strategies." OECD.AI Policy Observatory, 2024. OECD.AI

General

Europe's AI Paradox: Regulating What You Don't Build