France's AI infrastructure plan is moving from announcement to production. Mistral's new 44-megawatt data center in Bruyères-le-Châtel is operational with 18,000 NVIDIA GB200 systems, the first node in a roadmap to reach 200 megawatts of compute capacity across Europe by 2027. The site, announced a year ago at NVIDIA GTC Paris at VivaTech, is the clearest signal yet that France's frontier-AI ambitions have left the slide deck.
Mistral is also working with French public investment bank Bpifrance, MGX and NVIDIA to expand Campus AI, a network of AI factories anchored by a planned 1.4-gigawatt facility — one of Europe's largest AI campuses on the drawing board. The scale matters: 1.4GW is roughly seven times the 200MW Mistral plans to operate company-wide by 2027, putting Campus AI in the same weight class as hyperscaler buildouts in Texas and Arizona.
The hardware backing it is NVIDIA Blackwell. Scaleway, a European public cloud provider, now offers Blackwell B300-SXM instances on demand, while Bull and Foxconn have announced production of NVIDIA Vera Rubin NVL72 systems in Europe — manufactured and tested at Foxconn's Czech Republic facilities, then assembled, integrated and validated at Bull's factory in Angers, France. Schneider Electric is separately working with NVIDIA on reference designs for gigawatt-scale AI factories.
Key facts
- 01Mistral's 44MW data center in Bruyères-le-Châtel is operational with 18,000 NVIDIA GB200 systems.
- 02Mistral targets 200MW of compute across Europe by 2027, anchored by a planned 1.4GW Campus AI facility.
- 03Bull and Foxconn will manufacture NVIDIA Vera Rubin NVL72 systems in the Czech Republic and Angers, France.
- 04Orange Business has scaled its Live Intelligence GenAI platform to more than 100,000 internal active users.
- 05LINAGORA's Luciole 1B, 8B and 23B models were pretrained on the Jean-Zay supercomputer via OpenLLM-France.
The investment frame behind all of this comes from France 2030 commitments, the 2025 AI Action Summit and this year's Choose France Summit, layered with private capital from Bpifrance and MGX. A consortium of eight French companies has also submitted a bid to host a European AI gigafactory in France, which would consolidate France's claim as the EU's primary AI infrastructure hub.
The model layer is filling in alongside the metal. NVIDIA's Nemotron Coalition — a group of open-model builders that includes Mistral as a founding member — is producing open frontier models, datasets and training playbooks. LINAGORA's Luciole 1B, 8B and 23B models were pretrained on Jean-Zay, one of Europe's most powerful supercomputers, in collaboration with CNRS/IDRIS under the OpenLLM-France project, and shipped on Hugging Face along with their pretraining datasets.
H Company, also in the coalition, is building Holotron, a family of computer-use agents on open Nemotron models that interact with software interfaces the way a human would — no APIs, no custom integrations. Pleias has worked with NVIDIA on Nemotron-Personas-France and Nemotron-Personas-Belgium, privacy-preserving synthetic persona datasets grounded in local demographics, and is training compact language models on open, well-documented datasets to ease EU AI Act compliance around data provenance.
“What we see now is a shift from building one isolated model to running continuous model infrastructure, where models train the next models, curate data, generate synthetic environments and verify reinforcement learning.”— Pierre-Carl Langlais, Chief Technology Officer, Pleias
On the deployment side, French enterprises are starting to show production volume rather than pilot metrics. Orange Business, the B2B subsidiary of Orange, scaled its Live Intelligence GenAI platform to more than 100,000 active internal users before opening it to external customers as a regional agentic AI offering with EU data residency. Sanofi is deploying AI agents across research, manufacturing, commercial operations, procurement and IT, and is working with startups Owkin and Biolevate on autonomous agents for drug discovery.
Stellantis is rolling out AI-enabled digital twins across its global manufacturing footprint. Dassault Systèmes is combining virtual twins with open models on its agentic 3DEXPERIENCE platform. TotalEnergies is building Pangea 5, a next-generation supercomputer with Dell Technologies and NVIDIA for seismic imaging and energy-sector simulation. L'Oréal is using its CreAltech platform to pair generative AI with 3D digital twins for content production at scale.
The skeptics' case isn't hard to assemble. Europe still trails the US and China badly on training-compute totals, and 200MW by 2027 — while real — is a fraction of what a single hyperscaler now brings online in a year. The 1.4GW Campus AI facility is a plan, not a pour, and gigawatt-scale projects routinely slip on permitting, grid interconnect and power-purchase agreements. Mistral's open-model strategy also has to keep pace commercially with closed-weights competitors that monetize directly through APIs.
Even with those caveats, the shape of the French AI stack is now legible end-to-end: sovereign compute from Mistral and Scaleway, hardware assembled in-country by Bull and Foxconn, open models from Mistral, LINAGORA, H Company and Pleias trained on Jean-Zay and Nemotron, and enterprise deployment at Sanofi, Stellantis, TotalEnergies, Orange and L'Oréal. That vertical integration — compute, models, agents, customers, all inside the EU regulatory perimeter — is the bet France is making against the US-China duopoly, and it's the bet European regulators have been waiting for someone to actually fund.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
