Mistral Lets Enterprises Build AI on Their Own Data

Enterprise AI projects frequently falter due to models lacking understanding of specific business contexts, as they are often trained on general internet data instead of internal knowledge.

French AI startup Mistral has introduced Mistral Forge, a platform designed to allow enterprises and governments to create custom AI models trained on their own unique data. This offering was announced at Nvidia GTC, an event highlighting AI advancements for businesses.

Mistral's specialized focus on the enterprise sector contrasts with rivals who have prioritized consumer adoption. The company is reportedly on course to achieve over $1 billion in annual recurring revenue this year.

Mistral Forge enables companies to train models from the ground up, a contrast to common methods like fine-tuning or retrieval augmented generation. This approach could lead to better performance with domain-specific or non-English data, offering greater control over model behavior.

The platform utilizes Mistral's open-weight AI models, including smaller ones like Mistral Small 4. Forge also provides access to forward-deployed engineers who assist customers in identifying and adapting data for model training and evaluation.

Early partners for Mistral Forge include Ericsson, the European Space Agency, Reply, DSO, HTX, and ASML. Key use cases identified by Mistral involve governments needing tailored language models, financial institutions with strict compliance, manufacturers requiring customization, and tech firms needing code-specific tuning.