Europe AI Supercomputers: 35 New Systems Across 23 Countries in One Year

Europe is building 35 new artificial intelligence (AI) high-performance computing (HPC) supercomputers across 23 countries this year, the continent's largest one-year expansion of supercomputing capacity on record. The systems will reach more than 3 million researchers once operational, spanning national supercomputing centres, dedicated AI factories, and university research institutions from Spain to Sweden.

Barcelona Supercomputing Center, Germany's HLRS and Fraunhofer, Italy's CINECA, and Sweden's NAISS are among the institutions building new capacity, with the buildout collectively reaching 800 AI exaflops deployed or announced since last year. The announcement detailing the full scope of the expansion came today at ISC High Performance 2026.

The scale of this expansion is the actual story for anyone tracking how seriously European governments are treating AI as national infrastructure rather than a private-sector trend. Italy's CINECA frames its new IT4LIA system as strengthening technological autonomy. Germany's HLRS calls HammerHAI the country's first AI factory, built specifically for secure, national-scale computing. Sweden's Mimer AI Factory is owned outright by the EuroHPC Joint Undertaking, the European Union's (EU) coordinating body for shared supercomputing investment.

What each country is actually building

The Barcelona Supercomputing Center's EuroHPC AI Factory, the first EuroHPC installation built specifically for AI rather than general scientific computing, will expand its MareNostrum 5 system to deliver roughly 20 exaflops of AI training performance and 33 exaflops of inference performance, supporting climate modelling, biotech research, and government AI services for a consortium spanning Spain, Portugal, and Türkiye.

Bavaria's Blue Swan platform, hosted at Friedrich-Alexander University in Erlangen and the Leibniz Supercomputing Centre, is purpose-built for an independent multimodal AI foundation model targeting health and robotics applications. Bavarian Minister of Science Markus Blume described the underlying GPU cluster as the largest at any German university, framing the project as meeting European technical standards independently rather than relying on infrastructure built elsewhere.

Italy's IT4LIA carries the largest single deployment in this announcement, with over 8,000 graphics processing units (GPUs) delivering 82 exaflops of AI training and 164 exaflops of inference performance, built through a partnership between EuroHPC, CINECA, Italy's Ministry of University and Research, and the Italian Cybersecurity Agency. Germany's HammerHAI, Sweden's Mimer AI Factory, and the other systems in this expansion bring the total well past anything Europe has deployed in a single year before.

What triggered Europe's sovereignty push

Every government quoted in today's announcement uses some version of the same language: autonomy, independence, national infrastructure, European standards. CINECA's Gabriella Scipione describes IT4LIA as strengthening Europe's technological autonomy. Bavaria's Blume frames Blue Swan as meeting European standards independently. The Barcelona Supercomputing Center's Mateo Valero Cortés talks about giving researchers tools that meet the world's most complex challenges on Europe's own terms.

That language has a specific, recent trigger. Five days before today's announcement, leaders at the Group of Seven (G7) summit in Évian spent their final day confronting a question that had become impossible to avoid: what happens to a country's AI capacity when a foreign supplier can switch it off overnight. The United States Commerce Department had ordered Anthropic on June 12 to suspend access to its two most advanced models for every user outside the United States (US), citing national security grounds. European leaders left Évian without a resolution, but with the clearest argument yet for building AI infrastructure they fully control.

Why sovereignty and dependency are the same purchase

That argument sits inside a buildout where a single chip architecture powers more than 90% of the continent's AI factory capacity. Sovereignty, in this context, means controlling where the computer physically sits, who operates it, and which researchers and companies get access. It does not yet extend to the underlying silicon, which still comes from one company regardless of which country hosts the data centre. This is the same tension running through Europe's broader push for sovereign AI capacity, where infrastructure independence and deep reliance on a single international hardware supplier have so far developed side by side rather than in conflict.

The quantum layer running underneath

A separate but connected development sits inside today's announcement: Barcelona Supercomputing Center, CINECA, Fraunhofer, and the Jülich Supercomputing Centre are now using the same software platform to connect quantum processors to their classical supercomputers, extending what the industry calls quantum-GPU supercomputing. Researchers at Jülich, working alongside engineers from the underlying hardware platform, simulated a universal 50-qubit quantum computer, the largest such simulation completed to date, on Germany's JUPITER supercomputer.

This quantum integration matters for the same reason the broader buildout does. Simulating quantum problems on classical hardware lets researchers test algorithms and applications years before physical quantum computers reach the scale needed to run them directly. Europe is positioning itself to be ready for a quantum computing transition it cannot yet fully build, the same logic driving the AI infrastructure investment more broadly: build the capacity to use a technology now, even where full technological independence remains years away.

What this buildout signals for the next year

The 35 systems announced today are not Europe's final answer to the global AI infrastructure race. They are evidence of a continent treating compute capacity the way it once treated energy infrastructure or rail networks: as a strategic asset too important to leave entirely to market forces or to a single foreign supplier's roadmap. The G7's unresolved debate over access and today's new systems point to the same open question from opposite ends.

Control over the AI layer matters less if the compute underneath it can still be cut off by someone else's decision, and that gap will not be visible in exaflop counts. It will be visible in what European researchers and companies build once the hardware is running, and whether the breakthroughs land inside Europe's own fintech, healthcare, and energy sectors or simply accelerate work that gets commercialised somewhere else.

Editor's note

Every piece published on The Bright Minded goes through careful verification, but mistakes can happen. If you spot an error, have additional information, or want to flag anything, write to rosalia@thebrightminded.com.