Mistral's Big Bet on Europe's AI Infrastructure

The French AI company Mistral is broadening its business with three central announcements. First, it is building out costly compute capacity in Europe, bringing together a group of large European companies and institutions that commit to purchasing compute over several years. Mistral aims to grow European compute capacity to 200 megawatts by the end of 2027 and up to one gigawatt by 2030; in March it took on $830 million in debt for a 44 MW data center. The core of the new financing model is European Compute Units (ECUs), long-term commitments for multi-year access to Mistral's infrastructure, with partners expected to commit for about five years and no early exit. Participants include Amadeus, ASML, Capgemini, Caisse des Dépôts, and CMA CGM, and Microsoft will also rent capacity from Mistral's European data centers. Second, enterprise customers can now choose whether their AI requests are processed in Europe or the USA, and Mistral plans a variant using exclusively Mistral-controlled infrastructure without large cloud providers; it also introduces a service tier with custom usage limits and contractually guaranteed availability, claiming to be the only European AI lab offering both region choice and guaranteed availability. Third, Mistral will offer open-weight models from other developers on its platform, starting with GLM-5.2 from Chinese lab Z.ai, which will run on the same infrastructure with the same regional and availability commitments. The company argues that modern AI applications require a mix of models and that customers need both control over infrastructure and model choice; strategically, its infrastructure may become more important than its own models, as running large models and agentic applications on-premises becomes harder.