Europe Legislated. China Built.
Europe wrote the most comprehensive AI regulation in the world and called it leadership. What it actually did was impose costs on companies that need to compete globally, without securing equivalent commitments from anyone else, while doing almost nothing to build the industrial base that would give those regulations any leverage.
The instinct behind the EU AI Act is sound. Requiring transparency in high-risk AI systems, mandating human oversight, prohibiting certain applications outright - these are reasonable positions, and someone was going to have to take them. The problem is the sequencing and the scope. You cannot govern a race you are not running. And you cannot set global standards from a position of dependency.
The coordination trap
Unilateral regulation in a global market does one thing reliably: it creates arbitrage. Companies building the most capable systems will build them where the rules are most permissive, deploy them where the market is largest, and offer a compliance-ready version to European customers - usually the same product with a thicker terms-of-service and a data center in Dublin.
European users and European companies end up choosing between the regulated version - capability-limited, sometimes awkwardly constrained - and whatever the rest of the world has access to. This was predictable. It happened with GDPR: a decade of compliance overhead for European tech companies, a generation of European startups burning engineering resources on consent banners instead of products - while the platforms GDPR was actually aimed at absorbed the fines as a cost of doing business and carried on.
The EU Commission knows this. The response has been to position the AI Act as a template the world will eventually adopt - the Brussels Effect, the argument that European market size forces global standards. That theory has real grounding in consumer product regulation. In software, where distribution is instantaneous and jurisdictional arbitrage costs almost nothing, it is much weaker.
To its credit, the EU built the diplomatic infrastructure: the Council of Europe AI Convention, the G7 Hiroshima Code of Conduct, the Paris AI Action Summit. The results speak for themselves. The Biden administration signed the Council of Europe Convention in September 2024 - then Trump revoked Biden’s entire AI executive framework on his first day in office and sent no one to sign the Paris declaration two weeks later. China joined neither. Nothing binding emerged with either of the two powers that actually determine where the global AI frontier moves. The EU built a process and got nothing out of it. That points to a leverage problem, and leverage comes from having something to bring to the table.
The thing Europe didn’t do
Writing rules and building capability are different activities, and Europe has invested heavily in the first while neglecting the second. The United States has a chaotic, often brutal environment for technology companies, but it produces them at scale. Talent, capital, and ambition cluster around a small number of places that have become self-sustaining. Europe produces world-class AI researchers - Hinton did his PhD at Cambridge, LeCun at the Sorbonne, DeepMind was founded in London - and then watches them take jobs in San Francisco or build companies that end up acquired by US giants.
EU member states have not compensated for this with serious industrial support. France has made the most credible attempt, with a national AI strategy and backing for companies like Mistral. But one country’s effort in a 27-member union is a rounding error against the scale of what US federal procurement and defense spending routinely deploy. Germany’s Mittelstand model, excellent in manufacturing, does not translate to the capital intensity and speed that software development requires.
The EU needs to back promising homegrown AI companies at a scale that matches the ambition of the rules it writes about them. At the moment the rules are running well ahead of everything else.
What China is actually doing
Europe keeps filing the wrong complaint. Alleging unfair subsidies and filing trade disputes captures maybe a third of why Chinese AI companies are competitive - and responding to that third while ignoring the rest is how you lose the argument.
China’s advantage runs through infrastructure, industrial clustering, and a planning horizon that most European governments cannot match. The Chinese government has spent two decades building the physical environment in which technology companies can operate at scale - data centers, 5G networks, logistics infrastructure, power generation. It has cultivated regional specialization: the Pearl River Delta for electronics manufacturing, the Yangtze River Delta as the country’s leading semiconductor and AI production hub, clusters around Beijing and Shanghai for software research. Turning an AI model into a physical product - a robot, a sensor, a piece of industrial equipment - is dramatically cheaper in this ecosystem than anywhere else in the world.
Proximity and depth explain the cost advantage better than subsidies do. Chinese AI hardware companies are cheaper because their supply chains, component manufacturers, logistics networks, and skilled labor pools all sit within a few hundred kilometers of each other. The government’s role was to build and sustain that environment; once it exists, it compounds without ongoing intervention. Europe would be better served studying this than dismissing it as cheating.
The talent question
China runs active programs to bring distinguished researchers home and attract global talent to Chinese institutions. The United States - despite its dysfunction in immigration - remains the default destination for the world’s most ambitious AI researchers, because of its universities, its salary levels, and because the most interesting large-scale work is concentrated there.
Europe is a training ground. Students arrive from around the world, develop real capabilities, and leave - for San Francisco, for London, for wherever the most interesting work and competitive compensation can be found together. A senior AI researcher at a leading European university earns a fraction of what the same person makes at a US lab. No visa reform closes a 5x salary gap.
Closing it requires European companies with the scale to pay competitive salaries, European research institutions with the compute access to run experiments that matter, and a genuine shift in how Europe thinks about attracting people rather than just educating them. The US and China both treat talent acquisition as a strategic priority. Europe treats it as something that will sort itself out.
What regulation is actually for
Governing high-risk AI matters, and the EU was right to try. But rules without industrial capacity are a tax on the companies willing to comply with them, while their competitors elsewhere face no equivalent burden.
Europe’s actual goal should be a seat at the table when the real decisions about AI get made - about safety standards, about data governance, about what systems run critical infrastructure. That requires bringing something to the table beyond a well-drafted directive. Global standards get set by countries too important to ignore, not by the most carefully worded text.
Serious safety standards need a serious AI sector behind them. A competitive European AI sector gives European safety positions actual weight. And safety thinking done early shapes the technology rather than just auditing it after the fact.
The sequencing matters. Build first. Then the rules mean something.