Here’s the thing about enterprise AI that nobody wants to admit: most of it fails because the models don’t actually understand how your company works. They’re trained on the internet, sure, but they’ve never seen your internal documents, your workflows, or the decades of institutional knowledge that makes your business tick. That’s the gap French AI startup Mistral just decided to exploit.

Mistral just announced Mistral Forge, a platform that lets enterprises build custom AI models trained on their own data. The announcement came at Nvidia’s annual tech conference, and it’s a calculated move for a company that’s deliberately stayed focused on corporate clients while OpenAI and Anthropic chased consumer adoption. CEO Arthur Mensch says the strategy is working, Mistral is on track to hit over $1 billion in annual recurring revenue this year.

So what makes Forge different? While competitors offer fine-tuning or retrieval augmented generation (basically layering your data on top of existing models), Mistral is letting companies actually retrain models from scratch. This matters because it could mean better handling of non-English or highly specialized data, more control over how your model behaves, and the ability to train agentic systems without depending on third-party providers. You’re not tied to whatever OpenAI decides to do with their API next quarter.

According to Mistral’s head of product, Elisa Salamanca, Forge comes with all the tooling you need to generate synthetic data pipelines. But here’s where it gets interesting: the company also deploys its own engineers directly into customer operations. This borrowed-from-IBM approach helps companies figure out what data they actually need and how to evaluate whether their models are working. It’s consulting wrapped in a software product.

Mistral’s already signed up some heavy-hitters: Ericsson, the European Space Agency, ASML (the Dutch chipmaker that led Mistral’s latest funding round), and various government and financial institutions. The early adopter list tells you what Mistral thinks Forge’s main appeal is, governments needing models that understand their language and culture, financial institutions drowning in compliance requirements, manufacturers with specific customization needs, and tech companies trying to tune models to their codebase.

The bigger picture here is that Mistral has found its lane. While everyone else fights for attention in the consumer chatbot wars, this company is building the infrastructure that enterprises actually need. They’re betting that companies would rather control their own AI destiny than rent access to someone else’s black box. Given that they’re potentially heading toward a billion dollars in revenue, that bet might be looking pretty smart right now.