French foundation model developer Mistral AI SAS is reportedly in talks to raise €3 billion, or about $3.5 billion, from investors.

Bloomberg today cited sources as saying that the round could value the company at €20 billion. That’s nearly double the valuation it received after its most recent raise in September. The €1.7 billion investment was led by ASML Holdings NV, the world’s top maker of advanced lithography machines.

Today’s report didn’t specify the investors who could participate in Mistral’s new raise. ASML might chip in given that late-stage funding rounds often include contributions from existing backers. Mistral’s investor roster also includes Nvidia Corp., Salesforce Ventures and several prominent venture capital firms.

The report that the company is raising new capital comes shortly after it announced plans to develop artificial intelligence products for industrial engineers. The product lineup will include, among other things, custom AI models. Mistral may be planning to use the new funding to support the development effort.

Mistral refers to the software suite that it’s developing as physics AI. According to the company, the technology will enable engineers to generate multiple variations of a product design quickly. Mistral is also working on tools that will make it possible to test design variations in simulations.

In May, the company bought a startup called Emmi that developed tools for building physics AI models. Such algorithms feature a different architecture than a standard large language model. They’re optimized to solve partial differential equations, functions that lend themselves well to describing complex physical phenomena. 

Mistral researchers have released multiple papers focused on physics AI. Some contain training datasets related to computational fluid dynamics, a field that concerns itself with the aerodynamic properties of cars and aircraft. An earlier Mistral paper explored ways of applying AI to fusion research.

It’s possible the company will open-source some of its planned physics AI models. Mistral has released more than a half-dozen open-weight algorithms to date. The newest addition to the lineup, Mistral Medium 3.5, made its debut in April. It features 128 billion parameters and can outperform models more than three times its size across some coding benchmarks.

Mistral generates revenue with a suite of paid cloud services powered by its open-source models. Mistral Vibe is an AI assistant that can summarize documents, generate code and troubleshoot factory equipment. It’s available alongside an AI agent development platform called Studio. Additionally, Mistral offers AI-optimized cloud infrastructure and tools for building custom foundation models.

The company’s reported fundraising push comes as its two top rivals, OpenAI Group PBC and Anthropic PBC, prepare to go public. Mistral will also face competition from a new market player called Prometheus Inc. that raised $12 billion in funding this week. The company was launched by Amazon.com Inc. founder Jeff Bezos last year to automate many of the hardware engineering tasks that Mistral is prioritizing with its physics AI initiative.

Image: Mistral