Europe bids for AI infrastructure while Nvidia logs another demand surge

Mistral hosted its first AI summit in Paris this week, and the subtext was less about models than about infrastructure sovereignty. European executives and attendees framed the gathering as evidence that the continent is finally building compute capacity and talent networks that can compete with U.S. and Chinese clusters. The pitch is pragmatic: if foundation models matter to industrial policy, Europe cannot rely on rented American GPUs forever.

The timing matters because the capital window for AI infrastructure remains wide open. Nvidia CEO Jensen Huang described current demand as "insane" in recent remarks, a one-word summary that captures the order backlog his firm continues to work through. That language suggests 2026 will look a lot like 2025 for the chipmaker, with hyperscalers and sovereign buyers lining up for H100 and successor architectures.

Europe's bid is late but not irrational. If compute is the choke point, then countries with industrial policy budgets and data-residency mandates have reason to build. Mistral is positioning itself as the local champion, the firm that can train competitive models on European infrastructure and keep the stack inside regulatory borders. Whether that model pencils out depends on how much governments are willing to spend and whether European cloud providers can close the efficiency gap with AWS and Azure.

Huang's "insane" tells you where the capital is flowing. Mistral's summit tells you where the second-order build-out is happening. The question for 2026 is whether sovereign AI infrastructure becomes a distributed reality or a subsidy sink. Both can be true.