Mistral AI SAS today disclosed that it has raised $830 million in debt financing from a group of banks.

The consortium included Bpifrance, BNP Paribas, Crédit Agricole CIB, HSBC, La Banque Postale, MUFG and Natixis CIB. Paris-based Mistral will use the funds to build a data center in a suburb of the French capital. The facility is expected to come online in the second half of the year.

Best known for its line of open-source large language models, Mistral plans to equip the data center with 13,800 of Nvidia Corp.’s GB300 chips. Each accelerator combines one central processing with two Blackwell Ultra graphics processing units. The Blackwell Ultra includes 208 billion transistors based on a four-nanometer process.

The company will use the data center to run inference workloads and train new AI models. It released its most advanced LLM, Mistral Large 3, in December. The algorithm features 675 billion parameters and includes optimizations that make it particularly well-suited for powering AI agents.

Mistral offers Mistral Large 3 alongside more than a dozen other open-source models. Some are LLMs, while others are optimized for narrow tasks such as extracting text from documents.

The company generates revenue by offering cloud services that can be used to turn its models into enterprise applications. One of the company’s newest paid offerings, Mistral Compute, is an AI-optimized infrastructure platform powered by Nvidia’s GB300 chip. It provides managed AI hardware and bare-metal servers that offer an expanded set of customization options. 

Mistral Compute also offers software tooling designed to ease customers’ LLM projects. According to Mistral, the platform provides access to the pre-training workflows it uses internally to build LLMs. Another feature speeds up the task of loading a newly trained LLM from its testing environment to production.

Other paid offerings span a wide range of use cases. 

A product called Forge provides tools for building custom foundation models. It can generate synthetic data for AI training projects, as well as measure models’ performance using a set of built-in testing tools. Additionally, Mistral has a presence in the chatbot and vibe coding categories with its Le Chat and Mistral Vibe services, respectively.

The company’s planned data center in France will provide 44 megawatts of computing capacity when it comes online later this year. It’s part of a broader initiative that will also see Mistral build an AI campus in Sweden. The latter project is expected to cost $1.2 billion.

Mistral hopes to have 200 megawatts of computing capacity by the end of 2027. That corresponds to one-fifth of a gigawatt. Last year, Nvidia Chief Executive Jensen Huang stated that building 1 gigawatt of computing capacity costs between $50 billion and $60 billion.

Image: Mistral