Mistral Launches Forge Platform to Let Enterprises Build Custom AI Models

Mistral, the French AI startup, unveiled Forge, a platform that enables enterprises and governments to train custom AI models using their own data. Announced at Nvidia's GTC conference, Forge offers a library of open-weight models, including the new Mistral Small 4, and provides forward‑deployed engineers to guide customers through data preparation, evaluation, and infrastructure choices. Early partners such as Ericsson, the European Space Agency, Reply, and Singapore’s DSO and HTX are already testing the service. Mistral aims to address the gap where off‑the‑shelf models trained on internet data fail to understand specific business contexts, positioning itself as a serious contender against OpenAI and Anthropic in the enterprise market.

Introducing Mistral Forge

Mistral, a French artificial‑intelligence startup, announced a new platform called Forge at Nvidia's GTC conference. The service is designed to let enterprises and government agencies build custom AI models that are trained on their own proprietary data rather than relying on models trained primarily on internet content. By allowing organizations to train models from scratch, Forge seeks to close the gap that often causes enterprise AI projects to underperform.

Why Custom Models Matter

According to Mistral, many AI deployments fail because the underlying models do not understand the specific language, workflows, and institutional knowledge of a business. Off‑the‑shelf models are typically trained on broad internet data, which can limit their effectiveness in highly regulated or domain‑specific environments. Forge gives companies direct control over both the data and the behavior of their AI systems, reducing dependence on third‑party providers and mitigating risks such as model deprecation or unexpected changes.

How Forge Works

Forge provides access to Mistral’s library of open‑weight models, including the recently introduced Mistral Small 4, which is positioned as a smaller, more adaptable model. Customers can select the model that best fits their needs and then train it using their own datasets. The platform includes tooling for generating synthetic data pipelines, building evaluation metrics, and managing the required infrastructure.

In addition to the software stack, Mistral offers a team of forward‑deployed engineers (FDEs) who embed with customer teams to help surface the right data, design appropriate evaluations, and ensure successful model training. This service model mirrors approaches used by established players such as IBM and Palantir.

Early adopters and use cases

Forge is already available to a range of partners, including telecommunications giant Ericsson, the European Space Agency, Italian consulting firm Reply, and Singapore’s DSO and HTX. The Dutch chipmaker ASML, which led Mistral’s Series C financing round, is also an early adopter.

Mistral’s chief revenue officer highlighted several primary use cases: governments that need models tailored to local language and culture; financial institutions with strict compliance requirements; manufacturers seeking domain‑specific customization; and technology companies that want models fine‑tuned to their own code bases.

Strategic positioning

While rivals such as OpenAI and Anthropic have focused heavily on consumer adoption, Mistral is doubling down on the enterprise segment. The company reports that it is on track to exceed $1 billion in annual recurring revenue this year, underscoring the financial relevance of its enterprise‑first strategy.

By offering a full stack—from open‑weight models to engineering support—Forge aims to give businesses the ability to create AI systems that truly understand their unique data, opening new opportunities for AI‑driven efficiency and innovation across a variety of industries.