Europe Needs a Real CERN for AI, Not Wordplay

AI Digital Sovereignty R&D Policy: Europe Needs a Real CERN for AI

Europe has been calling for a "CERN for AI" for nearly a decade. The idea is simple: create a multinational research centre with the commitment, scale and structure to restore European leadership in a critical technology. The European Commission says it has delivered something "akin to a CERN for AI." It hasn't.

AI is now foundational infrastructure across science, industry and public services. Europe has world-class researchers, yet much of the technology they advance is built and deployed elsewhere. If Europe wants strategic autonomy in AI, it needs more than projects and press releases. It needs an institution built to last.

What the Commission announced - and why it falls short

Recent initiatives - branded as "akin to a CERN for AI" - include "AI Gigafactories," a network of supercomputers across Europe, and "AI for Science" calls under Horizon Europe. These are useful pieces, but they don't add up to a CERN for AI. They lack the legal foundation, independence, governance and long-term funding that made CERN work.

Without those elements, Europe will keep outsourcing core AI capability to non-European actors and remain dependent on external platforms and models. That is a sovereignty risk and a missed economic opportunity.

What made CERN work

There is a proven blueprint. CAIRNE (the Confederation of Laboratories for Artificial Intelligence Research in Europe) highlights six attributes behind CERN's success:

• Treaty-based foundation: An intergovernmental convention confers political legitimacy and stable financing.
• Independence: No direct control by industry, single nations or single organisations; funding is public to prevent private capture.
• Balanced governance: Each member state appoints both a scientific and a political delegate to the governing body.
• Substantial, long-term funding: Commitments extend beyond electoral cycles, enabling multi-decade projects.
• Physical concentration: Talent, compute and infrastructure in one place to create critical mass for hard technical challenges.
• Circulation of expertise: Researchers gather, collaborate, then return to home institutions and companies - spreading knowledge and networks across Europe.

These principles made CERN a global reference point in particle physics and even sparked breakthroughs like the World Wide Web. They are as relevant now as they were in 1954.

For context on how this model was set up and sustained, see CERN's official overview. And for the supercomputing backbone that already exists in Europe, see the EuroHPC Joint Undertaking.

What a real "CERN for AI" would require

• A multilateral treaty: Create a new intergovernmental organisation dedicated to AI research, evaluation and foundational infrastructure.
• Public funding only: Industry collaborates through open calls, testbeds and consortia, but cannot direct the agenda.
• Dual-delegate governance: Every member state appoints one scientific and one political representative to reconcile ambition with accountability.
• Decade-scale budgets: Lock in commitments sufficient to build and operate leading compute, data, evaluation and safety facilities.
• One primary campus, distributed satellites: Concentrate core capabilities in a flagship site, with linked national nodes for domain-specific programs.
• Mobility programs: Fund fellowships and exchanges so researchers rotate through and carry expertise back to their labs and companies.

Why this matters for science and research leaders

Without a treaty-grade institution, Europe will continue to trail in foundational model R&D, evaluation standards and critical data/compute access. That weakens bargaining position, slows translation into industry, and fragments efforts across national programs. A proper CERN for AI would reduce duplication, set common benchmarks, and provide shared infrastructure at a scale individual countries cannot achieve.

Practical steps to move from talk to action

• Form a member-state coalition: Assemble a core group willing to draft and sign a convention, then invite broader accession.
• Define a narrow, high-impact mandate: Focus on foundational models, evaluation and safety, open science infrastructure, and talent programs.
• Commit real compute and data: Pool and expand EuroHPC-grade resources, with guaranteed researcher access and transparent allocation.
• Stand up interim structures: Launch a legal pre-entity to scope governance, site selection and budget, while aligning existing national labs.
• Lock in mobility and training: Fund fellowships, sabbaticals and technical residencies tied to shared research roadmaps.

Bottom line

Europe does not need another label on existing programs. It needs an institution built with the same seriousness that created CERN. That means a treaty, independence, long-term money, a central site and a plan to circulate expertise back into every member state.

If the goal is AI sovereignty, a coalition of EU member states and like-minded countries should commit to a real CERN for AI - in name, structure and scale - and get the convention process underway now.