Open Models Drive Sovereign AI Spending

This marked the moment sovereign AI stopped being a policy slogan and became a buying program. China showed that frontier grade reasoning models could be released as open weights and run outside US labs, while Washington showed allies how quickly access to leading US systems could become a geopolitical lever. That combination turned AI from a procurement preference into a national infrastructure priority, like cloud, telecom, or payments rails.

• DeepSeek R1 mattered because it made the technical path legible. Governments no longer had to believe only OpenAI or Anthropic could build strong reasoning models. Open models meant a country could fine tune, host, audit, and restrict access inside its own borders or on a trusted regional stack.
  1 sacra 2 sacra 4 arxiv
• The demand pull came from politics, not just model quality. Europe had already been moving toward local control, but the sharper US posture accelerated budget release and vendor adoption. Mistral translated that into private deployments, on premise support, and sovereign compute, reaching $400M ARR by January 2026.
  1 sacra 3 sacra 5 sacra
• The winners are not necessarily the best pure research labs. They are the companies that can package models with local hosting, implementation help, and compliant infrastructure. That is why Mistral and Cohere gained traction, while weaker national champions risked sliding into consulting instead of owning the model layer.
  3 sacra 5 sacra 6 sacra

From here, sovereign AI spending will keep shifting from one off GPU purchases toward full national stacks, model, cloud, data residency, and deployment services bundled together. The durable leaders will be the labs and infrastructure providers that can deliver strong enough models with local control, because governments are now buying autonomy, not just intelligence.