The U.S. government’s decision this month to impose sweeping export controls on the most advanced Anthropic models, Mythos 5 and Fable 5, made Europe’s dependence on leading U.S. artificial intelligence providers starkly clear. Even if the U.S. government were to lift these restrictions soon, it is obvious that this can happen again at any time. Similarly, the growing scarcity of AI computing power makes political intervention to prioritize the needs of U.S. users a constant possibility.
Without a change of course, Europe risks becoming a tech backwater cut off from the most advanced capabilities, with possibly disastrous consequences for its security and prosperity. The recent “Europe 2031” scenario by leading European AI researchers and investors very convincingly depicts such a future.
The U.S. government’s decision this month to impose sweeping export controls on the most advanced Anthropic models, Mythos 5 and Fable 5, made Europe’s dependence on leading U.S. artificial intelligence providers starkly clear. Even if the U.S. government were to lift these restrictions soon, it is obvious that this can happen again at any time. Similarly, the growing scarcity of AI computing power makes political intervention to prioritize the needs of U.S. users a constant possibility.
Without a change of course, Europe risks becoming a tech backwater cut off from the most advanced capabilities, with possibly disastrous consequences for its security and prosperity. The recent “Europe 2031” scenario by leading European AI researchers and investors very convincingly depicts such a future.
Unfortunately, the most prominent of the initiatives for greater European independence, EuroStack, provides a blueprint that is at the same time unrealistic and not ambitious enough to deal with the possibility of near-term powerful AI. If things go wrong, its approach will not yield some partial success but a dangerously exposed and ultimately fully dependent Europe. Paradoxically, the most realistic path toward greater European ability to act is to build close relations with U.S. industry leaders of the current technology paradigm while doubling down on European strengths on industrial AI, placing bets on alternative technology paths, and building leverage by working together with other middle powers.
Any European AI strategy needs to account for deep uncertainty about the future trajectory of technology. It is entirely possible that AI progress under the current paradigm of large language models, or LLMs, will stall. If so, today’s frontier labs might even implode in spectacular fashion. But if the gigantic financial bet on the current paradigm pays off (with almost $700 billion invested alone in 2026) and what Anthropic CEO Dario Amodei calls “powerful AI” can be achieved in the coming years, it is absolutely essential that Europe secures access to the leading U.S. models. Second-best models would likely fall short of effectively guarding against critical cyber- and other security risks.
We received a taste of this scenario with the release of Anthropic’s Mythos 5 this year. Some of the rhetoric around its capabilities might have been hyperbolic, but the U.K. AI Security Institute’s evaluations did confirm that Anthropic’s highly guarded initial release was not just an elaborate charade. The lesson for Europeans must be that having the most advanced AI regulation does nothing to protect you against such risks if you don’t have access to the most advanced models yourself. In this scenario, access to a large amount of computing power is also likely necessary to avoid major economic disadvantages.
The EuroStack approach does little to nothing to deal with this possible future. With the motto “Buy European, Sell European, and Fund European,” its vision is to nurture European providers from applications to chips and data centers, with public procurement as a key lever. In its advocacy, the initiative builds on a rather grotesque overestimation of European capabilities. It dismisses the characterization of the European Union as a middle power and instead claims that Europe is a “SUPER power and we need to act like one” on AI policy.
Reality begs to differ. Europe’s best LLM, Mistral, currently ranks far behind the U.S. frontier models (and also the best Chinese) in terms of capability. Even if Europe were to throw maximum financial resources at Mistral right now, it is unlikely that the firm could close the gap with the top models where Elon Musk and Mark Zuckerberg have failed to date despite mind-boggling resource mobilization.
Indeed, EuroStack proponents acknowledge that “Europe is not going to build large frontier models but we can still build models a few steps behind which will be useful.” That still seems optimistic, given that European AI labs would be vulnerable to being cut off from U.S. hardware and compute infrastructure—unless Europe also manages to build a frontier chip designer and completely change its game on data center buildout at the same time. Europe currently only has 5 percent of the world’s computing power and is falling further behind. Publicly led investments into AI gigafactories have suffered delays and ultimately draw on meager EU funds, while European industry players see greater promise in less compute-hungry approaches to applied AI and largely refuse to fund data center investment at scale.
Rather than banking on AI superpower delusions, Europe should assume its role as a middle power and focus on deepening distinctive strengths, building leverage together with other middle powers and experimenting with alternative pathways. The goal should be to increase the costs for the United States to withhold access to LLM frontier models from Europe and to increase the incentives for U.S. companies to push for European access while simultaneously placing bets that will pay off if alternative technology paths prove successful.
One way to achieve this is to cement European strategic indispensability in the global AI ecosystem. Europe already has real strengths and assets: for example, ASML’s lithography machines, without which no high-performance chips can be manufactured, and Siemens Energy gas turbines, without which rapid data center expansion in the United States is barely possible. A key reason why they currently translate into little leverage is because Europe remains so existentially dependent on the United States in the military domain, making it vital to provide for its own security as fast as possible.
In managing the critical relationship with the United States, EU members need to build joint leverage and coordinate strategies with like-minded middle powers with significant capabilities, such as Canada, India, Japan, South Korea, and the United Kingdom. This cooperation can also put middle powers on a stronger footing vis-à-vis both AI great powers—the United States and China—including with a view to the urgently needed international agreements on managing catastrophic AI risks.
Moreover, it can serve to pool resources and capabilities where useful, including on building up state capacity. For example, the U.K. AI Security Institute provides a model for others to follow, including Germany, which has just decided to build its own version.
For expanding European compute capacity, publicly funded gigafactories can be an element serving core state and research needs, but enabling private investment will be key. As a recent study by the Carnegie Endowment for International Peace shows, the single-most important factor here is fast permitting processes and rapid grid connections. Investments should be drawn from Europe itself (such as Germany’s Schwarz Group), from partners including Japan (such as the recent 75 billion euro SoftBank investment in France), but also from the United States as a central part of the mix. Europe should invite U.S. hyperscalers and consortia facing growing public opposition at home to build data centers on the continent. In turn, firms will have every incentive to advocate for European access to the models that run on this infrastructure. Given the uncertainty of returns on massive data center investments, bringing U.S. investors on board also reduces European risk.
Finally, Europe must focus on preserving and reinventing its industrial strengths, both to build leverage and to capture a greater share of AI-enabled economic value. European policy should focus on building distinctive capabilities in areas of plausible competitive advantage, such as industrial AI and robotics. These areas are also closely linked to alternative technology paradigms such as world models and embodied learning loops, opening up opportunities for leapfrogging in a best-case scenario. To enable such advances, European regulation must support rather than hinder leadership in industrial AI. It is a positive step that the EU amended its AI Act to treat industrial AI systems differently from those for broad consumer use. It needs to do the same for industrial data in the EU Data Act.
To capitalize on its industrial base, Europe must avoid a selloff of proprietary data assets and promising innovations to foreign buyers. This clearly requires large-scale mobilization of European capital, including through capital markets integration and pension system reform. Moreover, European leaders need to make a real case for why Europe is the best place to build technology that could fundamentally be built anywhere. Comparatively predictable, rules-based policy is increasingly becoming an asset in its own right, but Europe must also build a much more promising home market by deepening single market integration for services and adopting a more solution-oriented approach to risk management.
At face value, this strategy may look less glamorous than the grand promises of complete European independence. But it is a much more realistic path toward securing European sovereignty in the sense of preserving and expanding Europe’s ability to defend its security and prosperity. It is also one that better positions the continent to seize the opportunities that the AI age presents.
—
Thorsten Benner is a co-founder and the director of the Global Public Policy Institute in Berlin.
Jakob Hensing is the head of political economy at the Global Public Policy Institute in Berlin.