Mistral wants open-weight AI to compete at the frontier. It just raised $3.5 billion to do it.

Mistral, a French AI company, recently announced it raised €3 billion in a Series D funding round, pushing its post-money valuation past €21 billion, according to a report. With this influx of capital, Mistral plans to expand its frontier research, scale compute capacity for model training, and grow its infrastructure. This move suggests that Mistral believes open-weight models can only achieve their potential if the compute and infrastructure underlying them are not concentrated among a few key players.

Open-weight models have the potential to democratize AI by allowing developers to choose from a wider range of models, reducing dependence on proprietary APIs. However, running powerful models necessitates massive amounts of compute. Training and serving frontier models require compute capacity concentrated among a relatively small number of labs, chip suppliers, and infrastructure providers.

Dario Amodei, CEO and co-founder of Anthropic, previously argued that open weights are not a sufficient solution as they merely shift the concentration to those with the most compute and chips.

Mistral aims to address this issue by building more than just open-weight models. The company is investing in the full stack required for AI solutions, including the infrastructure and compute capacity needed for model training and deployment, as well as the downstream products that bring these models into production. By developing an open-weight model stack, Mistral aims to free customers from dependence on a single vendor's roadmap, pricing, and availability, allowing them to build on Mistral's stack "without exposing their most valuable data, workflows, and institutional knowledge to anyone outside their walls."

This strategy is not new for Mistral, which has been releasing open-weight models since its launch three years ago. The company has also been expanding into the infrastructure layer, recently beginning to host third-party open models on the same infrastructure as its own models. In July, Mistral's CEO, Arthur Mensch, expressed concerns about dependence on closed-model providers, stating that enterprises need to use open-source models to avoid being forced into data retention by these providers.

Mistral believes that the company that ultimately dominates the AI landscape will need more than just the best-performing model; they must also control enough of the surrounding infrastructure to provide customers with choices about which models to use and on what infrastructure.