Mistral Launches Forge Platform to Let Enterprises Train AI From Scratch

Mistral, the French artificial intelligence startup valued at roughly €11.7 billion, has unveiled a new platform called Mistral Forge that enables enterprises and governments to build custom AI models trained entirely on their own proprietary data — a strategy the company hopes will differentiate it from rivals OpenAI and Anthropic in the lucrative corporate AI market.

Announced Monday at Nvidia's annual GTC conference, Forge goes beyond the fine-tuning and retrieval-augmented generation (RAG) approaches that dominate enterprise AI today. Rather than layering company data on top of pre-existing models, Mistral says Forge allows organizations to train models from the ground up using its library of open-weight models, including the recently released Mistral Small 4. Co-founder and chief technologist Timothée Lacroix told TechCrunch that the approach lets customers decide what capabilities to emphasize and what to deprioritize — a flexibility that smaller, general-purpose models inherently lack. The platform also includes synthetic data pipeline tools and comes with forward-deployed engineers who embed directly with customers, a consulting-style approach reminiscent of Palantir and IBM.

The enterprise bet appears to be paying off. CEO Arthur Mensch said Mistral is on track to surpass $1 billion in annual recurring revenue this year, a milestone that would represent extraordinary growth for a company founded just three years ago. While OpenAI and Anthropic have captured consumer mindshare with chatbot products, Mistral has deliberately concentrated on corporate clients — and Forge represents a deepening of that commitment. Early adopters include Dutch semiconductor equipment giant ASML, which led Mistral's Series C round last September; telecom leader Ericsson; the European Space Agency; Italian consultancy Reply; and Singapore's defense research organizations DSO and HTX.

Mistral's chief revenue officer Marjorie Janiewicz identified four primary segments for Forge: governments seeking models tailored to specific languages and cultural contexts; financial institutions with strict compliance requirements; manufacturers needing domain-specific customization; and technology companies looking to tune models to their proprietary codebases. The platform's emphasis on data sovereignty and organizational control could prove particularly attractive in Europe and Asia, where regulatory frameworks like the EU AI Act are imposing stricter requirements on how AI systems handle sensitive data.

The launch positions Mistral squarely against a growing field of enterprise AI contenders, but with a distinctive argument: that the persistent failure of corporate AI projects stems not from inadequate technology but from models that fundamentally don't understand the businesses deploying them. Whether training from scratch proves meaningfully superior to fine-tuning at scale remains to be seen, but Mistral is betting its billion-dollar trajectory on the answer being yes.