French startup Mistral unveiled an enterprise platform called Forge that lets customers build and train AI models on their proprietary data, aiming to wrest share from OpenAI and Anthropic. Announced at Nvidia’s GTC conference, the offering goes beyond fine-tuning and RAG by supporting training from scratch, giving clients greater control over behavior, language coverage and compliance, executives said. Chief Executive Arthur Mensch said Mistral is focused squarely on corporate buyers and is on track to surpass $1 billion in annual recurring revenue this year. Early users include Ericsson, the European Space Agency, Singapore’s DSO and HTX, consulting firm Reply and Dutch chipmaker ASML, which led Mistral’s Series C last September at a €11.7 billion valuation. The platform taps Mistral’s open‑weight models, including smaller variants that can be specialized for narrow domains, and pairs tooling for synthetic data and evaluation with forward‑deployed engineers. The push highlights intensifying competition to supply customizable AI stacks to governments, manufacturers and financial firms, even as questions remain about the costs and complexity of bespoke training versus more common adaptation techniques.
Related articles:
— Introducing Meta Llama 3
— Introducing Command R and Command R+
— Vertex AI: Build and Customize Generative AI Models
