Many enterprise AI initiatives falter not due to a lack of technology, but because the models employed fail to grasp the nuances of the business. These models are frequently trained using internet data instead of leveraging decades of internal documents, workflows, and institutional expertise.
Mistral, a French AI startup, aims to bridge this gap. On Tuesday, the company introduced Mistral Forge, a platform designed for enterprises to create custom models using their own data. This announcement was made at Nvidia GTC, Nvidia’s annual tech conference, where AI and agentic models for businesses are a major focus.
Mistral’s strategy highlights its commitment to corporate clients, contrasting with competitors like OpenAI and Anthropic, who have gained traction with consumers. CEO Arthur Mensch notes that Mistral’s dedication to enterprise clients is proving successful, as the company is projected to achieve over $1 billion in annual recurring revenue this year.
A key aspect of enhancing their enterprise focus is empowering companies with greater control over their data and AI systems, according to Mistral.
“Forge enables enterprises and governments to tailor AI models to their specific needs,” explained Elisa Salamanca, Mistral’s head of product, to JS.
While several companies claim to offer similar enterprise AI capabilities, most concentrate on fine-tuning existing models or integrating proprietary data via techniques like retrieval augmented generation (RAG). These methods typically do not involve complete model retraining but rather adapt models at runtime using company data.
Mistral, however, intends to allow businesses to build models from the ground up. This approach could potentially overcome some limitations of conventional methods, such as improved handling of non-English or highly specialized data, and enhanced control over model performance. This strategy could also empower companies to develop agentic systems through reinforcement learning and minimize dependence on third-party model providers, thereby mitigating risks like model updates or discontinuation.
Techcrunch event
San Francisco, CA
|
October 13-15, 2026
Forge clients can develop custom models utilizing Mistral’s extensive library of open-weight AI models, including smaller models like the recently launched Mistral Small 4. Timothée Lacroix, Mistral co-founder and chief technologist, stated that Forge can help extract more value from existing models.
“The compromises made when building smaller models mean they might not excel in every area as larger models do, so the ability to customize them allows us to decide what to prioritize and what to omit,” Lacroix noted.
Mistral provides guidance on model and infrastructure choices, but decisions remain with the customer, Lacroix emphasized. For teams requiring more than just advice, Forge includes Mistral’s forward-deployed engineers, who work closely with clients to identify the right data and adapt to their requirements—a model inspired by IBM and Palantir.
“Forge is equipped with all the necessary tools and infrastructure to create synthetic data pipelines,” Salamanca said. “However, knowing how to construct the appropriate evals and ensuring the right data volume is something enterprises often lack expertise in, and that’s where the FDEs add value.”
Mistral has already introduced Forge to partners, including Ericsson, the European Space Agency, Italian consulting company Reply, and Singapore’s DSO and HTX. Early adopters also feature ASML, the Dutch chipmaker that spearheaded Mistral’s Series C funding round last September, boasting a €11.7 billion valuation (around $13.8 billion at the time).
These collaborations exemplify Mistral’s anticipated primary use cases for Forge. According to Mistral’s chief revenue officer Marjorie Janiewicz, these include governments needing model customization for language and culture, financial institutions with strict compliance demands, manufacturers requiring tailored solutions, and tech firms needing model adjustments for their code base.
