Airbus for AI

Executive summary

What the proposal argues

Middle powers cannot match US frontier labs alone. A transnational public AI company — modeled on Airbus — is the structural answer. These six claims are the spine of that argument.

1. The problem is structural, not incidental. A handful of US corporations decide which governments can access frontier AI, what languages it speaks, and whose laws it follows. The compute, capital, and talent pipelines all run one way.
2. National responses don’t scale. No single European or middle-power country has the capital pool, talent depth, or compute footprint to compete with US labs. Sovereign-AI projects keep arriving late, underfunded, and trained on too little.
3. A consortium does. Pool the existing labs (Mistral, Cohere, Aleph Alpha, DeepL, Helsing, Salamandra, and others) into one entity. Give it anchor labs, regional contributors, and shared sovereign compute. Run it like Airbus: a transnational, public-private vehicle owned by member states.
4. Three tiers, three jobs. Tier 1 anchor labs do the frontier training. Tier 2 regional labs adapt models to local languages, law, and sectors. Tier 3 is shared sovereign compute (EuroHPC plus AI Factory sites) so none of the work has to live on AWS, Azure, or GCP by default.
5. Financing is the lever. Three vehicles already exist as templates: NATO Innovation Fund (defense-aligned), EuroHPC JU (compute), and a member-state cap-table (commercial). $20–30B over five years is the rough order of magnitude. That is achievable.
6. It can start now. A 10-person team and three months gets you a minimum viable entity. Then the Airbus parallel: anchor lab + first joint model + first sovereign compute MOU within 18 months.

Tan, Jackson, Berjon, Coyle. Airbus for AI: A global strategy for public value creation. Bennett School of Public Policy, September 2025.