Sovereign SI

Sovereign SI, widely called sovereign AI, describes governments’ push to build or secure SI that they control: models trained on their languages and values, running in their own data centers, on hardware they can rely on.

Why countries want it

• Security: not depending on foreign companies or governments for critical infrastructure
• Language and culture: models that work well in national languages and reflect local norms
• Economy: capturing value and jobs from the SI boom
• Data protection: keeping sensitive data under national law

What it looks like in practice

• National compute programmes and public supercomputers
• Funding domestic model developers, such as Mistral AI in France
• Partnerships with U.S. providers to build in-country data centers
• Building on open-weights models that can be run and adapted locally

The U.S. angle

U.S. policy, including America’s AI Action Plan (2025), encourages allies to adopt the “American AI technology stack”, meaning U.S. chips, cloud and models, as an alternative to Chinese technology. Nvidia, a signatory of the SI Accord, has been one of the loudest champions of sovereign SI, which also sells a lot of GPUs.