Mistral AI: From FluidStack Setback to Debt Financing Amid Market Uncertainty

Fluidstack is set to build, for Mistral AI, the largest AI cluster in Europe, comprising 18,000 GPUs.

The UK-based neocloud had announced this in March 2025. The cluster would be housed in a 40 MW Eclairion datacenter at Bruyères-le-Châtel (Essonne). The first phase was expected to be operational by the summer of 2025. The plan was to eventually reach a capacity of 100 MW.

One year later, the announcement had vanished from Fluidstack’s website (redirected to the homepage of the blog; it can still be found in the archive). Behind this, a shift toward the North American market. By the end of 2025, shortly after formalizing a $50B contract with Anthropic, the company announced its intention to relocate its headquarters to New York – though it has not yet done so.

Read also: Mistral AI launches Forge, an enterprise AI model training platform

In this context, Fluidstack would have abandoned the Bruyères-le-Châtel datacenter, where it was supposed to supply and operate the GPUs. It would, in parallel, abandon another large-scale project that France had showcased at the AI Summit in February 2025. Specifically, a 1 GW AI supercomputer in the Somme. The first phase was to rely on an initial investment of €10 billion, with an operational entry planned for 2026.

« Delivery doesn’t happen in a day »

Recently asked about the Bruyères-le-Châtel datacenter, Timothée Lacroix, CTO and cofounder of Mistral AI, did not mention Fluidstack. “It’s a big project, so delivery doesn’t happen in a day,” he said, in English. And he added: “Logistics and timelines are very different from what I usually encounter in software and research.It’s long-term planning.

According to Timothée Lacroix, the construction has nonetheless “progressed well.” “We are currently stabilizing the first tranche,” he explained. “We have a few jobs running and we’re making the final adjustments to ensure speed and stability.

Mistral AI will use part of this capacity for its own training needs. But it will also provide managed Kubernetes and Slurm stacks as part of the Mistral Compute offering, launched in June 2025 at VivaTech.

The unicorn has not, for the moment, faced major energy-access hurdles in Europe, according to its CTO. He does acknowledge some “constraints”: “The grid, in certain parts of Europe, is not easily scalable. I know this is a problem in France.” …

$830M borrowed, 13,800 GPUs expected

The remote-track of Fluidstack aside, Mistral AI is raising debt to buy GPUs itself. Specifically, 13,800 GB300 GPUs, for a capacity announced at 44 MW. It is borrowing $830M from Bpifrance, BNP Paribas, Crédit Agricole CIB, HSBC, La Banque Postale, MUFG and Natixis CIB.

Read also: Accenture and Mistral AI team up around sovereign generative AI

A commitment: the Bruyères-le-Châtel site will enter operational phase by at least June. And a longer-term objective: secure 200 MW of capacity in Europe by the end of 2027. In the foreground, its investment in Sweden: €1.2B for the construction of a datacenter in Borlänge, in partnership with EcoDataCenter. Opening planned in 2027.

HSBC is among Mistral AI’s clients. The two companies formalized their contract in December 2025. They do not reveal much beyond that the work will focus on the analysis of financial documents, multilingual reasoning (translation and processing of information) and productivity (“AI products” for marketing, procurement and customer service).

Bpifrance has already participated in several rounds of equity for Mistral AI. Including the latest (Series C), announced in September 2025, for €1.7B.

Dawn Liphardt

Dawn Liphardt

I'm Dawn Liphardt, the founder and lead writer of this publication. With a background in philosophy and a deep interest in the social impact of technology, I started this platform to explore how innovation shapes — and sometimes disrupts — the world we live in. My work focuses on critical, human-centered storytelling at the frontier of artificial intelligence and emerging tech.