A Blueprint for Multinational Advanced AI Development

Here's the link to learn more about this report: https://parispeaceforum.org/app/uploads/2025/11/a-blueprint-for-multinational-ai-development.pdf

Key findings:

• The U.S. controls ~75% of global AI compute capacity, China 15%, and the EU just 5%, creating an unprecedented concentration that threatens the sovereignty of mid-sized economies
• States face a choice between adopting foreign AI systems (risking dependency and coercion) or falling behind technologically (risking economic and security weakness)
• By pooling compute infrastructure, talent, and data, "AI bridge powers" can feasibly develop competitive frontier models while preserving sovereignty

The opportunity:

• Compute: EU AI Factories and planned "Gigafactories" represent €20+ billion in coordinated capacity deployable at frontier scale
• Talent: 87 of the 100 most-cited AI researchers originate from or work in countries outside the U.S. and China
• Strategic focus: Recent models like DeepSeek and Mistral show that architectural innovation and strategic choices can achieve near-frontier performance at dramatically lower cost
  

Why now:

"Current frontier training costs remain within reach of coordinated bridge powers, but the window is closing," the memo states. "Latecomers risk lasting strategic weakness given increasing barriers to entry, including compute monopolies, talent drain, and entrenched geopolitical leverage."

About this initiative:

The memo was developed by: 

• Mila - Quebec Artificial Intelligence Institute (Canada)
• The Oxford Martin AI Governance Initiative (United Kingdom)
• The Centre pour la Sécurité de l’IA (CeSIA) (France)
• The Technical University of Munich (TUM) (Germany)
• Uppsala Universitet (Sweden)
• RWTH Aachen University (Germany)
• The AI & Society Institute (France)