An Open Letter to European Heads of State
A Treaty-Based Framework for Sovereign European AI Capability
The Case
The most capable AI is being developed almost exclusively by a handful of private American and Chinese corporations, and Europe depends on nearly all of it. Europe has one frontier AI laboratory — Mistral, in Paris. Of the 36 AI models that had crossed the 10²⁵ FLOP training threshold between GPT-4 in March 2023 and October 2025, 28 came from US labs, six from China, and two from Europe. The combined AI capital expenditure of Microsoft, Google, Amazon, and Meta exceeded $200 billion in 2024 and is guided at roughly $725 billion for 2026. The EU’s InvestAI programme mobilises €200 billion — of which only €20 billion is public money.
The European Commission has responded with real ambition: nineteen AI Factories, a call opened in July 2026 for up to seven AI Gigafactories backed by up to €10 billion in EU and national funding against at least €20 billion in private investment, and an AI Continent Action Plan that promises to triple data centre capacity within seven years. But three independent analyses — by CEPS, Interface, and the Bertelsmann Stiftung — converge on the same diagnosis. The Gigafactories lack anchor customers. They are geographically dispersed for political balance rather than concentrated where energy is cheapest and talent is deepest. And they depend almost entirely on NVIDIA’s proprietary technology stack, potentially deepening the dependencies they were created to reduce.
The EU has built a compute layer. It has not built the research organisation that gives compute a purpose. Building AI Factories without a research institution is like constructing a particle accelerator without a physics department.
Three converging factors create a narrow window of opportunity. The AI architecture is in flux: state-space models and hybrid approaches may be more compute-efficient than Transformers, favouring innovation over brute-force scaling. Energy has become the binding constraint on AI progress, and Northern Europe has abundant, cheap, clean electricity. And the geopolitical landscape — US export controls, China’s acceleration, NATO’s northern expansion — makes sovereign European AI capability a strategic necessity rather than a luxury.
The gap is structural, not uniform. Europe leads on energy (8/10), where Northern Scandinavia’s renewable abundance is a genuine strategic advantage. But on compute and capital, the deficit is severe: Europe scores just 2/10 and 3/10 respectively. Those are the dimensions where concentrated institutional investment — exactly what Europa Nova proposes — can shift the trajectory.
The Proposal
An independent, treaty-based European research institution for sovereign capability in frontier AI. Modelled on CERN, but with a commercial licensing arm. Established by treaty among five to eight nations.
Pillar I
Governed by nine internationally recruited scientists. Led by a Scientific Director with veto authority on all technical decisions and market-rate compensation. One concentrated research hub in a major European city — not distributed across capitals. Full publication freedom for fundamental research. The research agenda is set by scientists, not by political bodies.
Pillar II
Located in Northern Scandinavia, running on 100% renewable hydropower at €0.03/kWh. Arctic cooling reduces energy costs by 30–60%. Initial capacity: 500MW, expandable to 2GW — at full utilisation roughly 4.4 and 17.5 TWh a year, against the 15 TWh that Statnett’s high scenario projects for every Norwegian data centre combined by 2040. This is a grid commitment, not only a construction project. Narvik was chosen for Stargate Norway for exactly these reasons — first by OpenAI, and after its withdrawal in April 2026, by Microsoft. Multi-vendor hardware strategy from day one — no lock-in to a single provider.
Pillar III
Wholly owned by the Foundation. Licenses models, tools, and compute access to European industry and the public sector. Revenue funds the mission. The mission generates the technology. Protected by international treaty, not corporate bylaws. Projected revenue: €500M by year five, rising to €2–3B annually by years eight to ten.
| Location | Elec. Cost | DC Share | Grid | Renewable |
|---|---|---|---|---|
| Established Hubs (FLAP-D) · 49–53°N — 62% of European DC capacity | ||||
| Frankfurt | €0.12/kWh | ~15% | Saturated | 52% |
| London | €0.10/kWh | ~14% | Constrained | 43% |
| Amsterdam | €0.09/kWh | ~13% | Moratorium | 40% |
| Paris | €0.08/kWh | ~11% | Moderate | 23% |
| Dublin | €0.11/kWh | ~9% | Critical | 36% |
| Nordic Emerging · 56–68°N — Fastest-growing region, lowest costs | ||||
| Norway | €0.03/kWh | ~4% | Abundant | 98% |
| Sweden | €0.04/kWh | ~3% | Abundant | 68% |
| Finland | €0.05/kWh | ~2% | Growing | 56% |
| Denmark | €0.06/kWh | ~2% | Planned | 84% |
The mismatch is geographic. Europe’s data centre capacity sits inside a four-degree band of latitude; the cheap, abundant, renewable power sits twelve degrees further north. FLAP-D hubs hold 62% of that capacity and face the highest energy costs and grid saturation. The Nordics offer electricity at one-third to one-quarter of the cost, with abundant grid capacity and near-total renewable supply. Microsoft, CoreWeave, and Brookfield have already committed over $15 billion to Nordic AI infrastructure.
Governance
Concentration over distribution — because dispersed resources produce dispersed mediocrity. CERN works because it is in Geneva.
Scientific autonomy. Political direction of research produces incremental work, so the Scientific Director controls the programme.
Market-rate compensation: 57% of the world’s top AI researchers already work in the United States. The alternative to paying market rates is not cheaper talent. It is no talent.
A sunset clause at year three, with independent evaluation and orderly wind-down if results fall short. This is what makes the proposal credible to sceptics.
No juste retour. Airbus’s decades of workshare distortions demonstrate the cost of proportional return.
Democratic oversight without operational control, modelled on central bank independence. The public has a right to know. It does not direct experiments.
Funding
Sovereign treaty contributions from five to eight nations provide approximately €3 billion per year. A 1–2% allocation from Norway’s Government Pension Fund Global — structured as infrastructure investment with residual value, not a grant — provides anchor capital. Commercial licensing revenue from year four progressively reduces the need for sovereign contributions. EIB co-financing covers physical infrastructure.
Context: OpenAI’s Stargate is $500 billion. Microsoft alone spent $80 billion on AI infrastructure in fiscal 2025. Europa Nova’s €5 billion per year is roughly four times CERN’s budget. It is the minimum credible stake in a contest where the buy-in rises every quarter.
Illustrative National Allocation
Total: €3 billion per year · €30 billion in treaty contributions over ten years · Each contribution is less than 0.06% of the nation’s annual budget
Roadmap
Questions and Objections
Policy Context
Recent developments in European AI policy, sovereign technology cooperation, and frontier AI governance — with commentary on their implications.
Endorsements
Express Support
Expressions of support are reviewed before publication. Submitting this form does not automatically add your name to the public signatory list.
You can also reach us directly at support@europa-nova.eu.