Mistral AI has raised 723 million euros through debt financing from seven banks, including BNP Paribas and Bpifrance. The funds will go toward a new data center near Paris that will be equipped with 13,800 Nvidia GB300 chips. With this, the French AI company is taking the next step in its ambition to build European AI infrastructure. With ASML as the primary investor, the goal is to eventually establish an AI powerhouse within Europe.
The data center is being built in Bruyères-le-Châtel, south of Paris, and is scheduled to go online in the second half of 2026. The 44 megawatts of computing capacity is intended for inference workloads and training new models. Notably, this is the first time Mistral has taken on debt for infrastructure expansion. Previous rounds were funded entirely by equity capital.
The 13,800 Nvidia GB300 chips are expected to deliver the AI performance within the facility. Each accelerator combines a single processor with two Blackwell Ultra GPUs, with each chip featuring 208 billion transistors built on TSMC’s 4nm process. Mistral is already incorporating the chips into its commercial offerings: the Mistral Compute platform provides customers with access to managed GPU hardware and bare-metal servers based on the same GB300.
From Lab to AI Cloud
Mistral’s infrastructure ambitions extend beyond this data center alone. Earlier this year, it acquired the French serverless cloud provider Koyeb to strengthen the aforementioned Compute platform; in Sweden, Mistral is building an AI campus expected to cost $1.2 billion. The company aims for 200 megawatts of computing capacity in Europe by the end of 2027.
Mistral has also secured a significant amount of funding from Veldhoven. In September 2025, ASML became the majority shareholder following an investment of 1.3 billion euros, bringing the company’s valuation to 10 billion euros. Previously, Mistral had already raised 600 million euros in June 2024 to finance its AI ambitions.
In addition to infrastructure investments, the product portfolio is also growing. Mistral Large 3, released in December, features 675 billion parameters and is specifically tailored to controlling AI agents. Additionally, Mistral offers tools for building custom foundation models via Forge and a chatbot service via Le Chat. The approach is significantly more complex than that of competing AI players in the U.S. While OpenAI and Anthropic focus primarily on developing AI models themselves and various applications for them, Mistral also aims to deploy its own capacity for a wider range of purposes. This is intended to prevent the French AI player from ending up in the same predicament as OpenAI, which is currently still searching for a profit model.