Europe Hosts Chinese Models Locally

Europe’s sovereign AI market is becoming an infrastructure and services business, not a model building race. The winner is increasingly the company that can host strong models inside European borders, connect them to a bank or factory’s internal systems, and give legal and security teams confidence that prompts, logs, and staff access stay local. Mistral’s switch to Z.ai’s GLM-5.3 inside Mistral Vibe shows that model origin now matters less than who controls deployment, GPUs, and customer relationships in Europe.1 2 3 4 5

• Mistral first won Europe by selling its own multilingual open weight models, then shifted after DeepSeek and other Chinese labs pulled ahead on cheap, high performance open models. By late 2026 it was using a Chinese model as the default in its own product while still keeping customer data and inference on European infrastructure.1 3 5
• What enterprises actually buy is not raw model access. They buy private deployments, GPU capacity, integration work, and compliance. That is why Mistral could scale to roughly $400M ARR serving groups like SAP, HSBC, BNP Paribas, Stellantis, and governments, and why European AI demand now looks a lot like cloud outsourcing with a local wrapper.1 2 4
• The closest comparable is Cohere, which competes for the same sovereign contracts from a different angle. Mistral is building and owning more of the stack, from data centers to orchestration, while Cohere focuses on enterprise software and runs on customer or partner infrastructure. The common thread is that sovereignty is shifting from who trained the model to who operates it locally.1 2 4

The next phase is a split between a few regional operators that own trusted AI infrastructure and a larger set of integrators that package foreign models for local industries. In Europe, that points toward sovereign AI looking less like a continental OpenAI and more like a regulated utility layer for running the best available models under European control.1 2 4