For years, critics of Europe’s AI strategy have made the same argument: while America builds artificial intelligence, Europe writes laws for it. The EU AI Act became the clearest expression of that divide — a landmark attempt to govern AI that many saw as further proof that Europe was prioritizing rules over innovation.
Recent events suggest the debate may be changing.
In June, U.S. restrictions temporarily blocked access to Anthropic’s advanced Fable 5 model, disrupting businesses and governments across Europe before the ban was lifted just weeks later. Around the same time, OpenAI came under growing pressure to limit access to some of its most advanced capabilities, while Anthropic itself reportedly faced political scrutiny over restrictions it placed on the use of its models for autonomous weapons and mass surveillance. Earlier this year, the Trump administration also reportedly threatened to reconsider government contracts with Anthropic after the company refused to allow its models to be used for certain military and surveillance applications.
Whether temporary or permanent, these episodes reveal a broader shift: frontier AI is no longer being treated merely as commercial software. It is becoming a strategic infrastructure.
That helps explain why Austria reportedly urged the European Union to explore attracting Anthropic to establish a stronger European presence. The proposal turned one of Europe’s greatest criticisms on its head. For years, the continent’s regulatory framework was portrayed as a reason frontier AI companies would avoid Europe. Now legal predictability, regulatory stability and the rule of law were being presented as competitive advantages.
The irony is difficult to ignore. The same AI Act which was critiqued that it might drive innovation away may become increasingly valuable in attracting AI companies.
But there is a deeper question hiding beneath the headlines.
Frontier AI doesn't grow on regulation tree
Even if Europe succeeds in becoming a safe harbor for frontier AI companies, how safe can that harbor really be?
A company relocating to Europe leaves behind the ecosystem on which frontier AI depends to operate and evolve. It still requires advanced semiconductors, hyperscale cloud providers, enormous computing capacity and vast amounts of affordable electricity. These are not peripheral inputs. They are the essential infrastructure of modern AI, and they remain overwhelmingly concentrated in the United States or under the control of American firms.
That dependence matters because it exposes the limits of what Europe can actually offer. A company may move its offices, employees and legal headquarters, yet it remains deeply dependent on an ecosystem of American cloud providers, semiconductor supply chains, compute infrastructure and, increasingly, the energy systems that power them.
More importantly, relocating a company does not eliminate political leverage.
Governments do not need to control every part of the AI value chain to influence its direction — they only need to control one critical link. Restrict access to advanced chips, cloud infrastructure, compute capacity or another essential dependency, and the entire strategy begins to unravel. In other words, moving a company does not necessarily move its dependencies.
This exposes a broader weakness in Europe’s AI strategy.
We need “Infrastructure Effect”
For more than two decades, Europe has exercised influence through what legal scholar Anu Bradford famously described as the “Brussels Effect”: the ability to shape global markets by setting the rules companies must follow. It worked remarkably well for privacy, consumer protection and competition policy because access to Europe’s market was itself a powerful incentive.
Artificial intelligence breaks this otherwise complex equation.
Unlike previous waves of digital innovation, frontier AI is constrained by height-end and difficult to replicate physical infrastructure.
Compute capacity, advanced semiconductors, hyperscale data centres, abundant electricity and deep pools of capital are no longer supporting assets — they are the new factors of production. They cannot be legislated into existence through regulation alone.
The numbers illustrate the scale of the challenge. Today, the United States controls roughly three-quarters of the world’s frontier AI compute, while Europe accounts for only a small fraction. American hyperscalers also power around 70% of Europe’s cloud market, and private AI investment in the United States is nearly ten times greater than in the European Union. At the same time, the U.S. accounts for almost half of global data-centre electricity consumption and continues to expand AI infrastructure at a pace Europe has yet to match. AI leadership is increasingly determined not only by who writes the rules, but by who owns the infrastructure.
Rules cannot replace compute
This is not the first time economic power has shifted because control over the factors of production changed. During the Industrial Revolution, governments could regulate commerce, but economic leadership ultimately belonged to those who controlled coal, steel, railways and factories. The AI revolution is creating a similar shift. Today’s strategic assets are GPUs instead of steam engines, hyperscale data centres instead of railways, and gigawatts of electricity instead of coal.
None of this diminishes the value of the AI Act. If anything, recent events suggest that Europe’s emphasis on legal certainty may prove to be one of its genuine competitive advantages. In a world of growing geopolitical uncertainty, predictable institutions matter.
But predictability is only one part of this much more complex equation.
The real risk for Europe is not that it chose to regulate AI. It is that regulation becomes a substitute for building the capabilities that ultimately determine who leads in AI.
Rules cannot replace compute. They cannot manufacture semiconductors, generate electricity, build hyperscale infrastructure or create industrial scale.
Europe may well become the preferred destination for companies seeking regulatory certainty. That would be a meaningful achievement.
But unless Europe can build the broader ecosystem to support the development of the frontier AI, it will have solved only part of the equation. Rules may attract frontier AI companies, but they cannot replace the capabilities those companies ultimately depend on.
Europe may be able to offer regulatory shelter.
But not yet strategic autonomy.
Member discussion