Mistral AI wants Europe’s biggest companies to pay now for AI capacity it plans to build over the rest of the decade. On Tuesday the French lab announced a coalition of enterprises making multi-year commitments that it says will underwrite 200 megawatts of infrastructure across Europe by the end of 2027 and a full gigawatt by the end of 2030, VentureBeat reported.
The commitments convert into “European Compute Units,” a claim on Mistral-built capacity that customers can spend on inference, training, model adaptation or raw compute. CTO Timothée Lacroix was blunt about the terms. “The goal is to have customers commit for around five years, or at least a long time,” he told VentureBeat. Asked about leaving early: “There is no getting out.”
The scale of the bet
Mistral currently runs less than 200 MW, per the company. Its near-term sites include a 44 MW facility near Paris that came online in the second quarter, a 23 MW site in Sweden built with EcoDataCenter and a 10 MW site in Les Ulis, France. A gigawatt is a different order of magnitude. VentureBeat cited Epoch AI’s estimate that a typical 1 GW AI data center needs about $38 billion in upfront capital, most of it for servers and GPUs.
That’s why the structure looks more like a power-purchase agreement than a cloud contract. Lenders want demand locked in before capital goes out, and Mistral has already raised €830 million in debt for its Paris-area data center. The anchor group includes Amadeus, ASML, Capgemini and shipping group CMA CGM. ASML, which led Mistral’s Series C funding round last year, called European AI capacity one of the few industrial efforts that “will matter more to Europe’s next generation.”
Sovereignty, with fine print
Alongside the compute units, Mistral made Regional Endpoints generally available, letting customers pin inference to Europe or the U.S., and put a Priority Tier with an uptime SLA into public preview. In-region inference is still subject to “limited, safeguarded transfers” to sub-processors outside the chosen region. Lacroix pointed to tool calls such as web search, which Mistral may need to gate for customers who require everything to stay in Europe.
The lab will also host third-party open models, starting with GLM-5.2 from China’s Z.ai. “It’s a great model. Everyone loves it. It’s open weight, so there was no good reason for us not to do it, really,” Lacroix said.
Microsoft is part of the financing picture too. Under a partnership expanded in July, it rents capacity in Mistral’s European data centers, which Lacroix compared to the neocloud model of selling capacity to hyperscalers.
The thesis behind all of it is that trillion-parameter models and agentic workloads will push enterprises off on-premises inference and onto clouds like Mistral’s. “I don’t see how, with the current trend of model size and growth of agentic tokens, we keep the full inference on-prem,” Lacroix said. Mistral has raised about $4 billion to date, according to PitchBook, a fraction of what OpenAI and Anthropic have, so those five-year contracts carry a lot of weight.


WebProNews is an iEntry Publication