As Europe faces being cut off from Anthropic’s new AI models, a group of European think tanks is arguing for an unorthodox path towards tech sovereignty: build more AI data centres with American tech giants.

The Trump administration on Friday ordered Anthropic, a leading US AI lab, not to provide its latest models to non-US nationals. The lab complied by disabling access for all customers just days after launching Fable, a more secure version of its much-discussed Mythos model.

Many European politicians proclaimed this to be yet another wake-up call and argued for greater support for European efforts to build the continent’s own AI models. But a group of AI experts across think tanks, academia and the investment community is instead doubling down on recommendations they released on Thursday as part of “Europe 2031”, a report detailing a fictional scenario for the continent’s future.

The authors are pushing for a massive buildout of AI compute in Europe and for alliances with other powers outside Washington and Beijing as a counterweight to the AI superpowers.

Throwing money at AI models

The Europe 2031 report does not paint a pretty picture. It argues that the US and China could dominate the world’s economy within five years by developing leading AI models, leaving Europe at the mercy of their whims and demands.

But the authors also argue that the answer should not simply be to throw public money at Mistral, Europe’s only horse in the race to build AI models. American AI companies are currently pouring billions of dollars into building AI data centres, training new models and hiring some of the most talented people on the planet. Some experts argue that it will be impossible for Europe to compete.

Juijn and the report’s other authors warn against the public sector trying to bankroll European AI, which they fear would waste money and political capital.

There’s also ASML, the Dutch company that dominates the production of the complex machines needed to fabricate leading-edge AI chips. Some experts argue that ASML could be Europe’s only leverage in the AI supply chain, but Juijn is sceptical.

“ASML is often called a choke point. I don’t think it actually is a choke point,” he said, arguing that if Europe restricted ASML’s exports, the US wouldn’t feel the effects immediately.

Build, baby, build

The authors instead urge Europe to massively increase the size of its AI computing capacities. The aim would be to draw in American giants, giving Europe some leverage over their future moves.

Without this base, they fear Europe’s push for sovereignty will leave it lagging technologically and without leverage if the political situation around AI becomes even more dicey a few years down the line.

This plan, however, would also require substantial funding. Europe would need to mobilise public and private capital at a scale it has “not attempted in peacetime”, the authors write.

The investment would also need to be backed by special zones for building this AI compute, similar to the “data centre acceleration zones” that the European Commission wants to establish through its Cloud and AI Development Act. That would require a targeted energy buildout and a radical streamlining of permitting procedures.

The China hope

The scenario is, of course, built on a central assumption: that American frontier AI models will grow ever more dominant. An alternative would be for Chinese open-source models, which have been close behind for some time, to provide a viable alternative.

“I should say that smart people disagree on this topic,” Juijn admits. But he argues that AI is increasingly being used to improve AI and that the runners-up will simply lack the money to build out as much AI compute as the leading labs.

And of course, adds Stan van Baarsen, another of the scenario’s authors, Chinese open-source models being a viable alternative depends on Beijing not deciding to restrict its own models for security reasons, just as the US is now already doing.

(cm, aw)