Mistral bets on ‘build-your-own AI’ as it takes on OpenAI, Anthropic in the enterprise

Mistral, the French AI startup, announced Mistral Forge at Nvidia's GTC conference, a platform designed for enterprises to build custom AI models trained entirely on their own proprietary data. This approach fundamentally differs from the industry-standard methods of fine-tuning existing models or using retrieval-augmented generation (RAG), which merely adapt models at runtime. Forge enables companies to train models from the ground up using their internal documents, workflows, and institutional knowledge, aiming to solve the core enterprise problem of generic models that don't understand specific business contexts. The platform leverages Mistral's library of open-weight models, like Mistral Small 4, and provides the tooling and infrastructure for generating synthetic data pipelines.

Mistral's strategy includes providing forward-deployed engineers (FDEs) who embed with customer teams to help identify the right data and build proper evaluations, a service model borrowed from companies like IBM and Palantir. Early adopters include Ericsson, the European Space Agency, and ASML, the Dutch chipmaker that led Mistral's recent €11.7B funding round. CEO Arthur Mensch states the company's enterprise focus is paying off, with Mistral on track to exceed $1 billion in annual recurring revenue. The move positions Mistral directly against OpenAI and Anthropic in the enterprise space, betting that giving companies more control over data and model training will unlock greater value than using generalized, internet-trained models.