Europe’s AI sovereignty push turns to hardware

The late-2026 burst of sovereign AI infrastructure activity did not emerge in a vacuum; it followed a sharp geopolitical shock that turned abstract dependence into an operational risk. As Computer Weekly reported, on 12 June 2026 the US Department of Commerce issued an emergency directive forcing Anthropic to suspend foreign nationals from using Claude Fable 5, with “no exemption for allied nations and no appeals process,” crystallising fears of a US “digital kill switch” and pushing European governments and businesses to build resilience into their technology stacks.

That is why the September-October moves mattered as acceleration rather than isolated announcements: they landed after sovereignty had already been translated into industrial policy and funding mechanisms. Computer Weekly noted that the EU Cloud and AI Development Act proposes “a single EU-wide sovereignty framework with four escalating levels of autonomy, from basic physical data residency (Level 1) through to full software supply chain” independence, while the UK’s “£1.1bn sovereign compute strategy,” announced by technology minister Liz Kendall, commits “£750m to a national AI supercomputer, including £400m for specialist chip procurement,” plus “£120m” for an AI hardware innovation programme.

Europe’s sovereignty push starts from a position of deep exposure to US-controlled digital infrastructure, which is why the response is taking the form of coordinated buildout rather than rhetorical autonomy. Cybernews reports that “Three American tech giants currently control approximately 70% of the European cloud market” and that “As many as 74% of publicly listed European companies use US-based services such as Google and Microsoft,” while fears over the US Cloud Act, a potential “kill switch,” and temporary limits on non-American access to advanced AI models have turned dependence into a security problem.

The answer emerging in Europe is to anchor AI capacity physically inside Europe, expand domestic industrial capability, and treat sovereignty as resilient interdependence built around secure, energy-aware infrastructure. According to the July 21 analysis, “Today, Europe hosts about 5% of the world’s compute. The US hosts about 75 to 80%,” which “means that we already are very dependent on the US for kind of European compute needs,” so the target is “around 15% of the global compute in 2031, which will be around 50 gigawatts” — a scale that ties strategic autonomy directly to data centers, power, security, and hardware access.

Europe’s sovereign stack is taking shape first at the processor-and-system layer, where SiPearl’s Rhea1 gives the region a domestically designed HPC CPU that is moving from concept into deployment. SiPearl said it “put the first Rhea1 into operation on 13 May 2026 and started a twelve-week functional bring-up,” with “test systems…handed over to partners and European research projects” after validation and “general availability…planned for the end of 2026,” while the chip is “to be used in the CPU cluster module of the JUPITER supercomputer at Forschungszentrum Jülich,” anchoring European processor design inside a European exascale architecture. Rhea1 is described as a “monolithic server processor with 80 Arm Neoverse V1 cores,” with per-core caches and “80 MiB of” shared cache.

What makes that stack sovereign is not autarky but control over the layers that matter most: architecture, memory design, deployment, and manufacturing locus. SiPearl says Rhea1 “combines four HBM2E stacks with a total of 64 GiB directly on the package and four DDR5 interfaces,” because the “distinguishing feature lies less in the raw core count than in the memory subsystem,” and adds that “European sovereignty here therefore does not mean complete autarky, but control over the specific process design, the security architecture, product development, and the intended use,” even as Fraunhofer’s Baltic Brain shows the same logic in AI training on owned hardware in Germany rather than US-operated cloud infrastructure.

The shift became visible in mid-September, when financing was immediately tied to buildout rather than left as a statement of intent. IO+ reported the headline “After €200M round, EUCLYD teams with FMC on sovereign AI,” and that pairing mattered: EUCLYD’s €200 million raise was not presented as abstract balance-sheet strength, but as capital linked to an industrial partnership with FMC, showing that sovereign AI plans were moving into execution with a manufacturing counterpart instead of remaining at the level of policy ambition.

Days later, the same pattern appeared deeper in the stack, as GlobeNewswire reported, “SiPearl Supplies First 80‑Core Rhea1 CPUs to Bull for Europe’s JUPITER Exascale Supercomputer.” That was followed in early October by Bull’s €80 million investment in its Angers plant, doubling capacity to build Europe’s sole AI supercomputers, so the sequence across late September and early October was unmistakable: European actors were no longer just announcing sovereignty goals, but delivering chips, pairing them with system integrators, and expanding factory capacity to turn those goals into production.