Mistral is betting on “build your own AI” as it competes with OpenAI and humanity in the enterprise

Most enterprise AI projects fail not because companies lack the technology, but because the models they use don’t understand their business. Models are often trained online, rather than through decades of internal documentation, workflows, and institutional knowledge.

This gap is where French AI startup Mistral sees opportunity. On Tuesday, the company announced Mistral Forge, a platform that allows companies to build custom models trained on their own data. Mistral announced the platform at Nvidia GTC, Nvidia’s annual technology conference, which this year is largely focused on artificial intelligence and agent models for enterprises.

It’s a clear move for Mistral, a company that has built its business on enterprise customers while competitors OpenAI and Anthropic have made significant inroads in terms of consumer adoption. CEO Arthur Mensch says Mistral’s laser focus on enterprise is working: The company is on track to exceed $1 billion in annual recurring revenue this year.

A big part of doubling down on enterprise efforts, says Mistral, is giving companies more control over their data and AI systems.

“What Forge does is it allows companies and governments to customize AI models to fit their specific needs,” Elisa Salamanca, head of product at Mistral, told TechCrunch.

Many companies in the enterprise AI space claim to already offer similar capabilities, but most focus on improving existing models or putting proprietary data on top through technologies like Retrieval Augmented Generation (RAG). These methods do not fundamentally retrain models; Instead, they adapt or query it at runtime using company data.

By contrast, Mistral says it enables companies to train models from scratch. In theory, this could address some of the limitations of more common methods – for example, better handling of non-English or domain-specific data, and greater control over model behavior. It can also allow companies to train agent systems using reinforcement learning and reduce reliance on third-party model providers, avoiding risks such as model changes or downtime.

Forge customers can build their own custom models using Mistral’s extensive library of open-weight AI models, which includes small models like the recently introduced Mistral Small 4. According to Mistral co-founder and chief technology officer Timothy Lacroix, Forge can help unlock more value from its existing models.

“The trade-offs we make when we build smaller models is that they can’t be as good at every topic as their larger counterparts, so the ability to customize them allows us to choose what we emphasize and what we drop,” Lacroix said.

Mistral advises on which models and infrastructure to use, but both decisions remain with the client, Lacroix said. And for teams that need more than just guidance, Forge comes with a Mistral team of future-deployed engineers who integrate directly with customers to surface the right data and adapt to their needs — a model borrowed from the likes of IBM and Palantir.

“As a product, Forge already comes with all the tools and infrastructure so you can create synthetic data pipelines,” Salamanca said. “But understanding how to build the right assessments and making sure you get the right amount of data is something that companies typically don’t have the right expertise for, and that’s what FDEs bring to the table.”

Mistral has already made Forge available to partners including Ericsson, the European Space Agency, Italian consulting firm RED, and Singaporean companies DSO and HTX. Also among the early adopters is ASML, the Dutch chipmaker that led Mistral’s Series C round last September at a valuation of €11.7 billion (about $13.8 billion at the time).

These partnerships are emblematic of what Mistral expects Forge’s key use cases to be. According to Marjorie Janewicz, chief revenue officer at Mistral, these include governments that need to design models that suit their language and culture; Financial players with high compliance requirements; Manufacturers with customization needs; And technology companies that need to adjust models to their code base.