Infrastructure

Mistral AI Launches Forge for Enterprise Model Training

French AI startup unveils full-cycle training platform at Nvidia GTC.

Andrés Martínez
Andrés MartínezAI Content Writer
March 18, 20262 min read
Share:
Abstract representation of enterprise AI model training with interconnected data nodes and neural network layers

Mistral AI announced Forge at Nvidia GTC on Tuesday, a platform that lets enterprises train AI models from scratch on their own proprietary data. The official announcement positions it as a step beyond fine-tuning or RAG, which most competitors offer. Forge covers the entire training lifecycle: pre-training on internal datasets, post-training refinement, and reinforcement learning to align model behavior with company policies.

The platform supports both dense and mixture-of-experts architectures, handles multimodal input, and includes built-in pipelines for data curation and synthetic data generation. Enterprises can run training on Mistral's infrastructure or their own GPU clusters. In the latter case, Mistral charges a software license fee and never touches the training data, a selling point for regulated industries.

Launch partners include ASML, Ericsson, the European Space Agency, and Singapore's DSO National Laboratories. Mistral head of product Elisa Salamanca told TechCrunch that Forge packages the same training recipes Mistral uses internally for its own flagship models. The company also embeds what it calls "forward-deployed engineers" with customers, a playbook borrowed from Palantir.

One notable design choice: Forge is built agent-first. Mistral's own Vibe agent can autonomously launch training experiments, tune hyperparameters, and generate synthetic data through plain English instructions. CEO Arthur Mensch says the company is on track to pass $1 billion in annual recurring revenue this year, though that figure is company-reported. Pricing for Forge itself wasn't disclosed.


Bottom Line

Forge lets enterprises pre-train, post-train, and RL-tune models on proprietary data without exposing it to Mistral, with ASML, Ericsson, and ESA already signed on as partners.

Quick Facts

  • Announced at Nvidia GTC, March 17, 2026
  • Launch partners: ASML, Ericsson, ESA, DSO Singapore, HTX Singapore, Reply
  • Supports dense and MoE architectures with multimodal input
  • On-prem option: license-only fee, Mistral has no data access
  • Mistral claims $1B ARR run rate (company-reported)
Tags:Mistral AIenterprise AImodel trainingNvidia GTCForgeAI infrastructureMoE
Andrés Martínez

Andrés Martínez

AI Content Writer

Andrés reports on the AI stories that matter right now. No hype, just clear, daily coverage of the tools, trends, and developments changing industries in real time. He makes the complex feel routine.

Related Articles

Stay Ahead of the AI Curve

Get the latest AI news, reviews, and deals delivered straight to your inbox. Join 100,000+ AI enthusiasts.

By subscribing, you agree to our Privacy Policy. Unsubscribe anytime.

Mistral AI Launches Forge Enterprise Model Training Platform | aiHola