March 17, 2026 — French artificial intelligence company Mistral AI is launching a new platform designed to let enterprises build AI models from the ground up using their own proprietary data. The move, announced at the Nvidia GTC conference, represents a strategic push to deepen its enterprise focus against rivals like OpenAI and Anthropic.
Addressing the Enterprise AI Gap
Mistral argues that many corporate AI projects fail because generic models trained on public internet data lack understanding of specific business contexts. The new platform, called Mistral Forge, aims to close that gap by enabling organizations to train models on internal documents, workflows, and institutional knowledge.
“What Forge does is it lets enterprises and governments customize AI models for their specific needs,” said Elisa Salamanca, Mistral’s head of product, in a statement.
This approach differs from common industry methods like fine-tuning or retrieval augmented generation (RAG), which adapt existing models rather than training new ones. Mistral claims training from scratch can better handle non-English languages, highly specialized domains, and offer greater control over model behavior.
Strategic Focus on Corporate Clients
While competitors have seen rapid consumer adoption, Mistral has built its business primarily on corporate and government clients. CEO Arthur Mensch stated the company is on track to surpass $1 billion in annual recurring revenue this year.
“The trade-offs that we make when we build smaller models is that they just cannot be as good on every topic as their larger counterparts, and so the ability to customize them lets us pick what we emphasize and what we drop,” explained Mistral co-founder and chief technologist Timothée Lacroix.
Forge allows customers to build custom models using Mistral’s library of open-weight AI models, including smaller options like Mistral Small 4. The company provides advisory support on model and infrastructure selection, but final decisions remain with the customer.
Deployment and Target Industries
For complex implementations, Mistral deploys teams of forward-deployed engineers to work directly with clients. This model, similar to approaches used by IBM and Palantir, helps organizations identify relevant data and build evaluation frameworks.
“Understanding how to build the right evals and making sure that you have the right amount of data is something that enterprises usually don’t have the right expertise for, and that’s what the FDEs bring to the table,” Salamanca noted.
Early partners and clients include Ericsson, the European Space Agency, Italian consulting firm Reply, and Singapore’s DSO and HTX. Dutch chipmaker ASML, which led Mistral’s Series C funding round in September 2025, is also an early adopter.
Primary Use Cases
According to Mistral’s chief revenue officer Marjorie Janiewicz, key applications include government agencies needing culturally and linguistically tailored models, financial institutions with strict compliance requirements, manufacturers requiring specialized customization, and technology companies tuning models to their specific codebases.
The platform includes tooling for generating synthetic data pipelines. This capability aims to help organizations overcome data scarcity challenges when training specialized models.
Market Context and Differentiation
The launch comes as enterprise AI competition intensifies. Mistral’s strategy emphasizes data control and reduced dependency on third-party model providers, addressing concerns about model changes, deprecation risks, and data sovereignty.
By enabling custom training, Mistral positions Forge as a solution for organizations seeking AI systems that reflect their unique operational knowledge rather than general internet patterns. The company’s European base may provide additional appeal for clients with strict data governance requirements under regulations like the EU AI Act.
Industry analysts note that successful enterprise AI adoption increasingly depends on integration with existing systems and proprietary knowledge bases. Mistral’s approach represents one attempt to solve this integration challenge through fundamental model retraining rather than surface-level adaptation.
This article was produced with AI assistance and reviewed by our editorial team for accuracy and quality.
