Europe faces hard choices on AI sovereignty

Europe’s dependence on American AI models is framed as a strategic vulnerability, but not necessarily as a permanent weakness. Access to foreign technology has historically supported economic convergence, and Europe has already benefited from major American software platforms such as Maps, Drive, Gmail, YouTube and Search. Broad export controls, such as the reported restriction affecting foreign access to Anthropic’s Fable model, may give way to more targeted American rules as model regulation becomes more sophisticated.

The economics of building a European frontier model look daunting. Meta is described as spending more on chips ($125 billion) than Germany spends on defense ($114 billion), while offering salaries of over $100 million to recruit top researchers, yet still struggling to catch up. Earlier European efforts to create a Google alternative, Quaero and Theseus, were discontinued by 2013. In a market where small capability gaps can produce large differences in outcomes, weaker models may struggle to generate revenue while trying to close the gap.

A “Buy European” approach risks locking public and private users into less capable systems if procurement favors local models over performance. Europe’s data center buildout also faces high power costs, dense populations and political risk, with operators warning that large-scale facilities may require subsidies. National governments retain major levers over competitiveness, including labor markets, taxes, permitting and energy policy, while European Union coordination may add complexity in any transatlantic dispute over model access.