Europe Opens Its AI Supercomputers to 2,300 Projects and Changes the Rules of the Game
32 AI computing centers operational by end of 2025, 145 million GPU hours allocated, 2,300 projects approved: Europe is transforming its promise of democratic access to supercomputing into measurable reality. EuroHPC Joint Undertaking has deployed its network of AI Factories and Antennas across the continent, creating a public alternative to American and Chinese technology giants.
The battle for European digital sovereignty is now being fought on the concrete ground of computing resource allocation. Between accelerated bureaucracy and the risk of capture by the best-organized players, the infrastructure exists, processes are being refined, and what remains to be proven is whether GPU hours actually translate into distributed innovation.
The Essentials
- 32 EuroHPC AI Factories and Antennas operational in 18 European countries by end of 2025
- 145 million computing hours allocated across 2,300 approved projects since launch
- Total budget of 7 billion euros over the 2021-2027 period for the EuroHPC program
- Three differentiated access modes: academic research, commercial development, and projects of European strategic interest
Infrastructure Catching Up to Political Promises
Europe took three years to transform its declarations of intent into operational capacities. The EuroHPC program, launched in 2018, had 32 computing centers dedicated to AI by the end of 2025, spread across 18 member countries. These AI Factories and Antennas offer access to exascale supercomputers such as JUPITER in Germany or LUMI in Finland, machines capable of 1,000 petaflops of computing power.
This physical scaling up is accompanied by an acceleration of project allocation processes. Project approval timelines have decreased from six months in 2023 to eight weeks in early 2025. EuroHPC has processed 4,200 access requests since 2023, with an acceptance rate of 55%. The 145 million hours allocated represents the equivalent of 16,000 years of computing on a standard processor.
The initiative is progressively catching up with the United States, where the National Strategic Computing Initiative has superior capacities but more restricted access. China maintains its quantitative advantage with more than 50 supercomputers in the global Top 500, but Europe is betting on openness to compensate for its volume disadvantage.
Three Speeds of Access That Reveal European Priorities
EuroHPC has structured its offering according to three distinct access modes that outline continental priorities. The “academic research” mode absorbs 60% of allocated hours, with projects involving climate modeling, drug discovery, or energy optimization. Technical University of Munich and France’s INRIA are among the largest consumers, with 8 and 6 million hours used in 2025, respectively.
The “commercial development” mode represents 25% of allocations. It allows European companies to access supercomputers for R&D projects, in exchange for a financial contribution indexed to their revenue. Airbus thus uses 3 million hours to optimize its aircraft designs, while German SAP consumes 2.5 million hours to develop its enterprise AI algorithms.
The remaining 15% is reserved for “projects of strategic European interest.” This category, created in 2024, funds initiatives deemed critical for digital sovereignty. It covers the training of multilingual European language models, digital twins of smart cities, or anti-missile defense simulations. These projects benefit from priority access and complementary funding that can reach 50 million euros per initiative.
France and Germany Dominate, Others Catch Up
The geography of the 2,300 approved projects reveals European imbalances in research capacities. Germany concentrates 28% of projects, primarily led by its technical universities and Fraunhofer centers. France follows with 22%, thanks to the mobilization of its public research organizations and engineering schools.
This Franco-German dominance is gradually fading. Italy has doubled its share of projects between 2024 and 2025, rising from 8% to 16% thanks to a national tax incentive policy for AI startups. Spain reaches 12% with support from its national artificial intelligence program, endowed with 600 million euros over five years.
Nordic countries outperform relative to their size. Finland, with its LUMI supercomputer, hosts 6% of European projects with only 1.1% of the continent’s population. Denmark and Sweden reach 4% and 5%, respectively, notably thanks to their deeptech ecosystems and early digital policies.
This geographic distribution directly influences application domains. Germany prioritizes automotive industry and engineering, France bets on aeronautics and defense, Italy develops augmented fashion and design, while Nordic countries excel in climatology and renewable energy.
Startups Access Computing, but GAFAMs Keep the Data
EuroHPC succeeded in its bet to open supercomputing to emerging European players. 400 of the 2,300 approved projects come from startups or small and medium-sized tech companies, compared to fewer than 50 in 2023. These companies access computing capacities that would cost them between 2 and 15 million euros annually on commercial platforms like Amazon Web Services or Microsoft Azure.
The impact is concrete. French startup Owkin uses 500,000 EuroHPC computing hours to train its medical AI models on European hospital data. German DeepL consumes 800,000 hours to improve its translation algorithms without transferring its linguistic data to the United States. Italian D-Orbit exploits 300,000 hours to simulate its earth observation space missions.
This democratization remains limited by the persistence of closed American ecosystems. While European startups access computing, they remain heavily dependent on AI frameworks developed by Google, Meta, or OpenAI. More problematic, 70% of training data used by EuroHPC projects remains hosted on American servers, due to the lack of European alternatives at the scale needed.
France attempted to bypass this bottleneck with its “AI data sovereignty” program, which funds the creation of European datasets in health, environment, and industry. Endowed with 200 million euros, it aims to reduce dependence on data controlled by American technology giants. Results are still too modest to change the game: only 12% of EuroHPC projects use exclusively European data.
Resource Allocation Reveals Geopolitical Tensions
The management of 145 million allocated hours illustrates European geopolitical tradeoffs in AI. EuroHPC created in 2024 a “strategic reserve” of 20 million hours, activatable in case of crisis or project of continental urgency. This reserve was mobilized for the first time in March 2025, during the Russian cyberattack against Ukrainian electrical infrastructure, allowing real-time modeling of energy bypass solutions.
Allocation criteria evolve according to international tensions. Since the escalation between Taiwan and mainland China, EuroHPC explicitly prioritizes projects for “technological resilience” against Asian supply chain disruptions. This priority benefited European research on alternative semiconductors and critical materials, which now represent 8% of allocated hours compared to 2% in 2024.
Growing climate concerns also influence allocation. EuroHPC reserves 15% of its hours for climate modeling and environmental adaptation projects, a proportion in constant increase since 2023. This green priority comes with energy efficiency requirements: EuroHPC computing centers operate with 85% renewable energy, compared to 40% for the average European datacenters.
The European initiative is gradually changing the rules of the global supercomputing game. It offers a credible alternative to American and Chinese monopolies, while revealing the limits of digital sovereignty that remains dependent on software ecosystems and data controlled by other powers. Between promises kept on infrastructure and persistent challenges on technological independence, Europe is building its computing capacity but is still seeking its path toward digital autonomy.