Mistral Forge: Train Custom AI From Scratch on Proprietary Data — Direct Challenge to OpenAI and Anthropic Fine-Tuning Paradigm

TechCrunch·medium signal

Mistral announced Forge at NVIDIA GTC 2026, an enterprise platform enabling companies to train custom AI models from scratch on their own data rather than fine-tuning a shared foundation model. This challenges the prevailing approach from OpenAI and Anthropic where enterprises get fine-tuning and retrieval on top of locked foundation weights. Mistral's thesis is that companies with sensitive data or highly specialized domains will pay a premium for full weight ownership.