145 million hours of high-performance computing. This is the volume that the European Union has made available to its researchers and businesses over the past five years via its supercomputers. A massive effort at pooling resources that contrasts with the continent’s traditional fragmentation and reveals an unprecedented strategy: transforming public infrastructure into a lever for technological sovereignty.
Europe is betting on shared computing to close its gap against American AI giants. With 19 AI Factories deployed across the continent, the Union is attempting to democratize access to computational resources on which innovation now depends. But will this public strategy be sufficient against the American private ecosystem that concentrates approximately 75% of global computing power?
The essentials
- 145 million computing hours allocated to 2,300 European projects via EuroHPC over 5 years
- 19 AI Factories deployed to pool access to high-performance computing
- The United States holds approximately 75% of global computing power, with Europe in a minority position
- 7 billion euros invested in European supercomputers for the 2021-2027 period
2,300 European projects powered by pooling
The EuroHPC Joint Undertaking initiative illustrates a radical shift in approach for Europe. Since 2018, this joint undertaking has financed and operated supercomputers shared between member states. The result: 2,300 research and development projects have benefited from these pooled computational resources, from climate modeling to industrial optimization.
This collaborative approach contrasts with the historical European model where each country developed its own infrastructure. France with its Jean Zay supercomputer, Germany with its distributed computing centers, Italy with its specialized facilities. EuroHPC now centralizes the allocation of resources according to scientific and economic needs, not according to national borders.
The 19 AI Factories constitute the logical extension of this strategy. These centers specialized in artificial intelligence offer democratized access to the most powerful GPUs, training data, and development tools. The objective: enable European SMEs to develop AI solutions without investing millions in infrastructure.
The catch-up strategy through public infrastructure
Europe accepts a minority position in the computing race. Faced with American domination that controls approximately 75% of global computing capacity, the European continent remains in a catch-up position. This asymmetry is not merely technical: it determines who can train the most advanced AI models and therefore who defines tomorrow’s technological standards.
The European response relies on the leverage effect of public investment. The 7 billion euros allocated to the EuroHPC program for the 2021-2027 period aims to create an ecosystem where private actors can experiment without bearing the infrastructure costs alone. This logic of technological common good draws inspiration from past successes: CERN for particle physics, Ariane for space.
Each AI Factory targets specific sectors. Finland specializes in climate applications, Ireland in digital health, Germany in Industry 4.0. This specialization avoids dispersing resources while maintaining continental coverage. The model recalls European ambitions in biotechnology, with shared infrastructure to compensate for geographic fragmentation.
The scale gap with American private giants
European efforts run into an arithmetic reality: the power gap with American private actors. Google, Microsoft, and Amazon each invest more than 10 billion dollars per year in their computing infrastructure. The 7 billion euros European effort spread over 7 years (2021-2027) appears modest compared to this financial firepower.
This asymmetry translates into concrete capabilities. The largest European clusters offer tens of thousands of GPUs. OpenAI or Meta computing farms align hundreds of thousands of specialized processors. When Europe pools resources for 2,300 projects, a single California company can mobilize the equivalent to train a language model.
The question is no longer merely budgetary. It becomes geopolitical. Access to high-performance computing determines who can develop the most powerful algorithms, process the largest volumes of data, offer the most sophisticated services. Without computational parity, Europe risks becoming a captive market for innovations designed across the Atlantic.
Promising results but uncertain industrialization
Early returns from European AI Factories encourage the continuation of the experiment. Supported projects benefit from an environment favorable to development, even if success metrics vary according to application sectors. The health, energy, and transport sectors particularly benefit from this democratized access to advanced computing.
The European meteorological institute was thus able to improve the accuracy of its climate forecasts by 23% thanks to EuroHPC resources. German SMEs optimized their production chains with average productivity gains of 15%. These isolated successes validate the technical approach.
But scaling to industrial levels remains problematic. Most supported projects remain at the prototype or demonstration stage. Europe excels in applied research but struggles to transform its innovations into commercial products capable of rivaling American or Chinese solutions. This industrialization difficulty is not specific to AI and reveals a structural European challenge.
The challenge of training in the face of talent shortage
Infrastructure alone is not enough: Europe desperately lacks specialists capable of exploiting these computational resources. AI Factories record more than 15,000 training requests for advanced AI tools, but can only meet 40% of them due to a lack of qualified instructors.
This skills shortage reveals a vicious circle. Without a critical mass of talent, European companies struggle to leverage access to high-performance computing. Without attractive industrial outlets, European engineers migrate to Silicon Valley or Asian R&D centers. This brain drain weakens the European ecosystem at the moment when the AI battle is being fought.
EuroHPC attempts to break this dynamic by coupling resource access with intensive training programs. Each AI Factory offers curricula of 6 to 12 months to train operational specialists. The objective: create a pool of 50,000 AI experts by 2030. An ambitious bet that will require convincing European universities to adapt their curricula to industrial needs.
A model that questions the effectiveness of pooling
The European approach to computing pooling raises fundamental strategic questions. Can one catch up with private giants using shared public infrastructure? Is the logic of technological common good suited to the speed of AI innovation?
Early results show the limits of the model. Despite 5 years of efforts and substantial investments, the gap with the United States continues to widen. European companies do gain access to computing resources, but their American competitors have dedicated infrastructure without allocation constraints or sharing requirements.
This tension between pooling and performance questions the future of European strategy. Should we persist in public logic or encourage the emergence of European private champions capable of investing on the scale of the GAFAM? Experience in other technological sectors suggests that a mixed approach might be necessary, combining public infrastructure and a dynamic private ecosystem.
Europe has 19 AI Factories and a budget of 7 billion euros to democratize access to high-performance computing for the 2021-2027 period. This pooled infrastructure allows 2,300 projects to develop without massive investments in computational resources. But the scale gap with American giants raises the question: can pooling rival the private concentration of resources? The answer will be determined in the next five years, when these public investments must transform into industrial innovations capable of standing up to Silicon Valley.