NVIDIA points to France’s ramp-up of “AI factories” and production-grade infrastructure

France is moving from announcing artificial intelligence ambitions to building production-grade infrastructure, with NVIDIA highlighting multiple projects coming online across compute, enterprise software and model ecosystems. In a post tied to VivaTech and NVIDIA’s broader European push, the company says “AI agents are running in production,” while startups and established enterprises are deploying applications and developing models, datasets and platforms geared toward local languages and European requirements.

A key example is Mistral’s new data center in Bruyères-le-Châtel in northern France. The project, announced at GTC Paris a year earlier, is now described as operational for its first deployment, using 18,000 NVIDIA GB200 systems. NVIDIA says this is intended to support a longer-term roadmap, with Mistral aiming for 200 megawatts of compute capacity across Europe by 2027.

NVIDIA links the center’s ramp to the design of its Blackwell platform, which the company says is built to help data centers maximize AI throughput under fixed power budgets. According to NVIDIA, that involves both higher performance-per-watt hardware and software features aimed at improving data center throughput in power-constrained environments, an increasingly central constraint for customers scaling training and inference.

The post also points to a larger compute campus concept in France. NVIDIA says it is working with Bpifrance, MGX and Mistral to expand Campus AI, described as a network of AI factories anchored by a planned 1.4-gigawatt facility. NVIDIA characterizes the initiative as one of the largest AI campuses in Europe, positioning it as a platform to concentrate capacity and accelerate deployments.

Beyond capacity buildouts, NVIDIA describes cloud and systems efforts aimed at giving developers and enterprises on-demand access to accelerated compute. Scaleway, a European cloud provider, is said to offer NVIDIA Blackwell B300-SXM instances. Separately, NVIDIA says Bull and Foxconn have announced production of NVIDIA Vera Rubin NVL72 in Europe, with early manufacturing and testing at Foxconn facilities in the Czech Republic before systems are assembled, integrated and validated at Bull’s factory in Angers, France.

NVIDIA also cites industrial and infrastructure partners to broaden the “AI factory” theme. Schneider Electric, the post says, has teamed with NVIDIA to develop blueprints for gigawatt-scale AI factories, aimed at helping organizations accelerate AI infrastructure deployment. The company also notes that a consortium of eight French companies has submitted a bid to host a European AI gigafactory in France, framed as a way to strengthen European AI infrastructure and speed adoption.

A recurring theme in NVIDIA’s update is that AI capabilities increasingly depend on systems of models and data, not just individual model releases. The post describes a shift toward continuous model infrastructure, in which models help train the next models, curate data, generate synthetic environments and verify reinforcement learning. It also emphasizes open model infrastructure as a mechanism to make AI more transparent and adaptable, with the goal of inspection, adaptation, deployment and auditing aligned to Europe’s compliance and trust requirements.

To illustrate what that looks like in practice, NVIDIA points to “open model” work discussed on stage at VivaTech with leaders from Gradium, H Company, LINAGORA, Pleias and NVIDIA. Pierre-Carl Langlais, chief technology officer of Pleias, is quoted saying open model infrastructure helps many people build AI and supports the dissemination of frontier-level practice throughout the economy. NVIDIA also points to Nemotron, describing it as a set of open models, datasets and playbooks intended to help model builders accelerate workflows from training through deployment.

NVIDIA’s examples extend beyond AI infrastructure to enterprise adoption. It says Sanofi is deploying AI agents across its value chain, from research and manufacturing to commercial operations and day-to-day functions like procurement and IT, and that it is working with startups Owkin and Biolevate on autonomous agents for drug discovery and development. NVIDIA also says Orange Business first tested and scaled its Live Intelligence GenAI platform internally, with more than 100,000 active users, before making it available across Europe as a trusted agentic AI solution with data hosted in the region. Additional references include Stellantis’ plans for AI-enabled digital twins, Dassault Systèmes’ use of its agentic 3DEXPERIENCE platform with virtual twins and open models, and TotalEnergies’ Pangea 5 supercomputer with Dell Technologies and NVIDIA.

It remains unclear from the update how much of the French buildout has been fully commissioned beyond the Mistral deployment, and NVIDIA does not provide independent performance benchmarks, cost details, or verified utilization rates for the broader set of initiatives. The post also does not break out which specific regulatory frameworks the described deployments are mapped to, or whether each project has been audited against particular compliance regimes. What is clear is the directional message, France is scaling compute capacity while pairing it with model ecosystems and enterprise deployment paths. Investors and industry watchers may look next for commissioning milestones at Campus AI, details on how large-scale power budgeting is managed as more AI factories come online, and evidence of sustained production workloads for the “agentic” systems NVIDIA highlights as moving beyond pilots.