Mistral AI has launched Forge, a platform that allows companies and governments to train custom AI models using their own proprietary data. The French AI startup announced the product at Nvidia’s GTC conference.
Most enterprise AI tools today rely on fine-tuning or retrieval augmented generation (RAG), which adapt existing models at runtime without retraining them. Mistral says Forge goes further by supporting the full training lifecycle, including pre-training on large internal datasets and reinforcement learning to align models with internal policies. The company claims this approach can handle problems that off-the-shelf models cannot, such as specialized languages, legacy codebases, or documents with unusual characters.
Early customers include Ericsson, ASML, the European Space Agency, and Singapore’s DSO. Mistral says it helped Ericsson build a model for translating legacy code, and assisted a public institution in reconstructing damaged ancient manuscripts.
Forge offers two deployment options: Training on Mistral’s own infrastructure or on the customer’s own servers, with Mistral never accessing the data.
The platform also comes with forward-deployed scientists who work directly with customers. Mistral CEO Arthur Mensch says the company is on track to exceed $1 billion in annual recurring revenue this year.
Mistral positions Forge against cloud providers such as Amazon, Microsoft, and Google, arguing that their tools lack depth and require customers to upload sensitive data.
Sources: TechCrunch, VentureBeat
