Mistral Forge: Train-From-Scratch Enterprise AI Announced at GTC — Forward-Deployed Engineers, Ericsson and ESA Already Live

TechCrunch·high signal

Mistral announced Forge at NVIDIA GTC — a platform enabling enterprises to train models entirely from scratch on proprietary data, bypassing fine-tuning limitations for non-English, domain-specific, and agentic RL use cases. Forge ships with Mistral forward-deployed engineers (Palantir-style delivery model) and is already live with Ericsson, European Space Agency, Reply, and Singapore's DSO/HTX. The announcement directly targets regulated industries and governments where foundation model data contamination is unacceptable.