AI in Europe: The Case Against Sovereign Models
Foreign citizens were banned from using Anthropic’s Fable model as the Trump administration utilized export controls to restrict access for all users, including Americans. According to industry reporting, this move demonstrates the U.S. government’s capacity to shut down frontier AI access through trade regulations, cementing a structural dependency for European users and businesses.
Why was Anthropic’s Fable model restricted?
The Trump administration used export controls to block foreign access to the Fable model because it provided the fastest mechanism to shut down access for everyone, including U.S. citizens. Reporting indicates that as American regulations evolve, these restrictions will likely become more narrowly targeted rather than broad bans.
This control mirrors the existing U.S. dominance in consumer software. European businesses and consumers have relied on Google Maps, Drive, and Search for two decades. While AI carries higher economic stakes, the qualitative nature of the dependence remains the same: Europe uses American tools to drive its internal economy.
Can Europe build a competitive sovereign AI?
Current capital requirements make a European “sovereign AI” effort unlikely to succeed. Meta plans to spend $125 billion on chips this year. To put that in perspective, that figure exceeds Germany’s entire national defense budget of $114 billion.
Past attempts at European digital sovereignty failed. In 2005, France and Germany attempted to build Quaero, a Google alternative. Disagreements led to a German spin-off called Theseus. Both projects were discontinued by 2013.
The financial gap is stark. Revenue reported by Mistral earlier this year represents less than one percent of Anthropic’s current revenue. Because AI returns are non-linear—where a tiny increase in performance leads to massive payoffs—second movers often earn nothing.
What is stopping AI data center growth in Europe?
European data centers cannot scale without subsidies, according to interviews with regional providers. High power prices and population density are primary hurdles. While smaller centers are appearing in Scandinavia and Iberia, the largest American investments are avoiding the EU.
Political risk is a deciding factor. American companies report concerns that the European Union will implement extractive taxes or change copyright rules after the capital is already committed. This lack of trust is compared to the energy transition, where European voters accepted the decline of the German car and chemical industries.
How should European nations respond to U.S. AI dominance?
Individual member states have more leverage than the European Commission. National governments control labor markets, taxes, and energy permitting. For example, the Netherlands could loosen planning laws to accelerate the expansion of ASML, while Germany could restart nuclear power to slash electricity costs.
Coordinating a unified EU response often increases costs and diminishes leverage. Certain member states, including Lithuania, Estonia, and Latvia, prioritize the U.S. military presence in Europe over AI independence, creating conflicted interests within the bloc.
If AI progress slows, the “increasing returns” advantage of U.S. labs may fade. Currently, Google sells millions of chips to Anthropic because Anthropic’s belief in near-term progress exceeds Google’s own internal projections. If this momentum stalls, European providers may find a window to catch up.
| Metric | U.S. Frontier (Meta/Anthropic) | European Efforts (Mistral/EU) |
|---|---|---|
| Annual Chip Spend | $125 Billion (Meta) | Significantly Lower |
| Revenue Share | Dominant Market Share | <1% of Anthropic’s Revenue (Mistral) |
| Strategy | Rapid Scaling / Capital Intensive | Sovereign AI / Subsidy Dependent |
Frequently Asked Questions
Why can’t Europe just use the “Buy European Tech Act”?
If “Buy European” means purchasing inferior models, it risks baking in security vulnerabilities and reducing productivity compared to frontier models.
Is dependence on U.S. AI permanent?
Not necessarily. Dependence on software does not imply permanent vassalage. New breakthroughs in fields like plasma physics or fusion reactors could create new industries where European firms lead.
What is the “intelligence commoditization” thesis?
It is the idea that worse models can substitute for frontier ones. However, experience with superior models like Opus 4.8 suggests that once users experience high-tier intelligence, lower-tier models become painful to use.
What do you think? Should Europe focus on building its own “Sovereign AI” or lean into its role as a major user of American technology? Share your thoughts in the comments or subscribe to our newsletter for more deep dives into the AI economy.