In 2026, the corporate AI race is heating up, but many projects still stumble for a simple reason: generic models don't understand a company's history. French AI firm Mistral is making a direct play to fix that.

At Nvidia's GTC conference this week, Mistral unveiled Mistral Forge, a system designed for enterprises to construct custom AI models trained entirely on their own documents, workflows, and data. This contrasts with the prevailing industry method of lightly adapting existing models. Mistral's approach involves training models from the ground up, which the company argues leads to better handling of niche jargon, non-English languages, and specific operational logic.

"Forge lets enterprises and governments customize AI for their specific needs," said Elisa Salamanca, Mistral's head of product. The platform provides the tools, but also offers a key service: teams of forward-deployed engineers who work on-site to help clients identify the right data and build effective training pipelines.

For Mistral, this is a strategic reinforcement of its enterprise-first stance, as competitors like OpenAI and Anthropic captured broader consumer attention. CEO Arthur Mensch claims the focus is working, with the company projected to exceed $1 billion in annual recurring revenue this year.

Early adopters of Forge include Ericsson, the European Space Agency, and chipmaker ASML, which led Mistral's latest funding round. According to Mistral executives, target clients are governments, financial institutions, and manufacturers—sectors where off-the-shelf AI often falls short. The bet is that true customization, not just fine-tuning, will be the differentiator that makes corporate AI finally work.

Source: TechCrunch