The European Strategy for Artificial Intelligence in Science starts from a blunt diagnosis: Europe was once ahead in scientific use of AI, but has been overtaken by the US and China, both in AI-intensive publications and in compute capacity, patents and commercial players. The EU’s share of AI compute is below 5%, compared to 75% for the US and 15% for China, with similarly modest shares of AI players and patents. The message is clear: without decisive intervention, Europe risks becoming a spectator in a world where scientific breakthroughs depend on large models, massive datasets and compute-hungry experiments.
The answer for the European Commission is RAISE (Resource for AI Science in Europe), presented as a virtual institute that pools four strategic resources across Member States and the private sector: compute, data, talent and funding. RAISE is supposed to serve two roles at once: (1) “science for AI” – pushing the frontiers of AI itself, with a strong emphasis on safe, robust and trustworthy frontier AI; and (2) “AI in science” – accelerating scientific discovery by embedding AI across disciplines, from life sciences and materials science to humanities and social sciences. The design is clearly inspired by global peers (NAIRR in the US, UK and Chinese initiatives), but with an explicit European twist: human rights, explainability, transparency and scientific integrity are framed as core design principles.
To get there, the Commission proposes a mix of targeted pilots and future structural funding. In the short term, RAISE gets a €107 million pilot under Horizon Europe, including €58 million for Networks of Excellence and Doctoral Networks, plus a pledge to dedicate €600 million to compute access for science via AI Gigafactories. Beyond that, the Commission signals a political ambition to double Horizon Europe’s annual investments in AI, exceeding €3 billion per year and explicitly boosting AI in science. Whether this survives negotiations on the next Multiannual Financial Framework is an open question.
Structurally, the Strategy tries to tackle three long-standing weaknesses in Europe’s research system: fragmentation, infrastructure gaps and talent competition. RAISE aims to knit together existing excellence into Thematic Networks and a European Network of Frontier AI Labs, which should reduce duplication and provide shared access to infrastructure. The communication stresses the need for long-term funding and access to EU-level compute and data that cannot be efficiently provided by Member States alone. In terms of talent, the focus is twofold: attracting and retaining world-class AI scientists and building AI skills across disciplines, with fellowships, doctoral networks and mobility schemes that spread expertise. Again, this relies on national systems being willing to support cross-border mobility and recognise European-level networks in their own career structures.
On data and ethics, the Strategy is ambitious in rhetoric but more cautious on concrete mechanisms. It promises support to identify strategic data gaps, curate and integrate “AI-ready” datasets for science, and develop domain-specific foundation models. It also commits to AI that is explainable, accountable, safe and aligned with European values, and explicitly ties AI in science to public trust and scientific integrity. However, it says comparatively little about governance specifics: Who controls access to high-risk models in sensitive research areas? How are biases and downstream harms assessed when scientific models are later repurposed in industry or public services? How are conflicts between open science and data protection / IP handled in practice? These are the issues where regional actors and ethics bodies will need to fill in the blanks.
The political positioning of the Strategy is important. It is not an isolated R&I document: it is framed as an implementing piece of the AI Continent Action Plan, and is adopted alongside the Apply AI Strategy. That means AI in science is part of a coordinated push: build scientific capacity (RAISE), accelerate deployment in industry and public sector (Apply AI), and maintain overall coherence and sovereignty (AI Continent plan, AI Act, Data Union Strategy). For Member States and regions like Flanders, this interlocking architecture has consequences: national AI-for-science programmes that ignore these links may find themselves sidelined when it comes to EU funding, infrastructure location and standard-setting.
From a Flemish perspective, the Strategy is both an opportunity and a warning. Flanders has strengths in several of the scientific domains (life sciences, materials, climate, language and vision technologies). If Flemish institutions move quickly to anchor themselves in RAISE Networks of Excellence, co-invest in compatible compute and data infrastructures, and align their own AI and data strategies with Apply AI’s sectoral priorities, they can punch above their weight in the next wave of AI-driven science. If not, RAISE risks consolidating excellence elsewhere, and Flemish actors may depend on external infrastructures with limited say over governance, access conditions or ethical frameworks.
Criticism of the EU’s AI-in-science strategy clusters around three points: ambition without backing, weak governance, and unresolved data and IP barriers. Observers broadly support the direction but argue that RAISE is simply too small and too slow to matter at a global scale, with pilot funding that pales next to US and Chinese investments and no guarantee of sustained budgets across future framework programmes. The “CERN for AI” branding is also questioned: instead of a strong, independent institution, RAISE risks becoming a loose, Brussels-managed network with unclear decision-making on which labs qualify, who controls access to compute and data, and how bureaucracy is avoided. Finally, the most concrete criticism targets copyright and data access, which the strategy largely skirts: paywalls, technical restrictions, fragmented text-and-data-mining rules, and an outdated commercial/non-commercial split are seen as actively undermining AI-driven research. The result, critics warn, is a strategy with the right narrative but insufficient money, muddled governance, and legal bottlenecks that threaten to block its own objectives.
