While the world keeps talking about the biggest models, Mistral is taking a different path. At Nvidia GTC, the French AI company unveiled Forge — a platform that lets enterprises train their own AI models. Not fine-tuning, not RAG, but actual training on proprietary data.
The Problem Forge Solves
Most enterprise AI projects don’t fail because of the technology. They fail because the models don’t understand the business. A GPT or Claude knows the internet — but not your internal documents, workflows, and the institutional knowledge built up over years. That’s exactly where Forge comes in.
How It Works
Companies can build custom models based on Mistral’s open-weight models (including the new Mistral Small 4). The platform supports pre-training, post-training, and reinforcement learning. For teams that need more hands-on support, Mistral sends its own engineers directly into the company — a model borrowed from the likes of Palantir and IBM.
Who’s Already On Board?
The customer list is impressive: ASML (the Dutch chipmaker that also led Mistral’s Series C), Ericsson, the European Space Agency, and Singapore’s defense organizations DSO and HTX.
There’s a specialized variant for the financial sector too: banks and insurers can train models on their compliance frameworks and regulatory documentation — and the data stays in-house.
A Billion Dollars in Revenue
CEO Arthur Mensch says Mistral is on track to surpass one billion dollars in annual revenue this year. For a company founded in 2023, that’s remarkable. The valuation stands at 11.7 billion euros since the Series C last September.
My Take
Mistral’s strategy is smart: instead of competing with OpenAI and Anthropic for the best general-purpose model, they’re becoming the platform provider for custom enterprise AI. It’s less glamorous than launching a new GPT or Claude — but it might be the better business. Once companies have built their own models on Forge, they’re not switching anytime soon.
Sources: