What happened
Mistral AI announced Forge at NVIDIA GTC on March 17, a platform that lets enterprises train custom AI models from scratch on their own proprietary data. Unlike competitors that rely on fine-tuning or retrieval-augmented generation, Forge supports full pre-training, post-training, and reinforcement learning so organizations can build domain-aware models that learn their vocabulary, reasoning patterns, and constraints. Launch partners include ASML, the European Space Agency, and DSO National Laboratories Singapore.
Why it matters
Forge represents a meaningful shift in the enterprise AI market. Instead of adapting a general-purpose model, organizations can now build models that deeply understand their domain from the ground up. This is particularly relevant for regulated industries (finance, defense, government) where data sovereignty and compliance requirements make hosted fine-tuning insufficient. Mistral reports it is on track to surpass $1 billion in annual recurring revenue, suggesting real enterprise demand for this approach.
Who should pay attention
- Enterprise architects evaluating build-vs-buy for domain-specific AI
- MLOps engineers managing model training pipelines at scale
- Teams in regulated industries (finance, healthcare, defense) needing data-sovereign models
- Organizations currently relying on RAG that want deeper domain integration