Most enterprise AI initiatives face hurdles not due to a lack of advanced technology, but primarily because the AI models employed often fail to grasp the specific intricacies and context of a company's business operations. These models are typically trained on vast amounts of internet data, lacking exposure to decades of internal documents, unique workflows, and proprietary institutional knowledge. It is this crucial gap that Mistral, the prominent French AI startup, aims to address with its new offering.

On Tuesday, at Nvidia's annual technology conference, Nvidia GTC—a major event this year focusing heavily on AI and agentic models for enterprise—Mistral unveiled Mistral Forge. This innovative platform is designed to empower enterprises to construct highly customized AI models, specifically trained using their own proprietary data. This launch represents a strategic move for Mistral, which has consistently prioritized corporate clients, differentiating itself from rivals like OpenAI and Anthropic, which have focused more on consumer adoption. CEO Arthur Mensch affirms this enterprise-centric approach is proving successful, with Mistral projected to exceed $1 billion in annual recurring revenue this year.

A core component of Mistral's commitment to the enterprise sector is providing businesses with enhanced control over their data and AI systems. Elisa Salamanca, Mistral’s head of product, emphasized that Forge enables enterprises and governments to tailor AI models precisely to their unique requirements. While several companies in the enterprise AI landscape claim to offer similar capabilities, most rely on fine-tuning existing models or integrating proprietary data via techniques such as retrieval augmented generation (RAG). These common methods adapt or query models during runtime and do not fundamentally retrain them.

In stark contrast, Mistral Forge distinguishes itself by allowing companies to train AI models entirely from scratch. This fundamental difference is poised to overcome limitations inherent in more conventional approaches. For instance, training from scratch can significantly improve the handling of non-English languages or highly domain-specific data. It also grants greater control over model behavior and facilitates the development of agentic systems through reinforcement learning, thereby reducing dependency on external third-party model providers and mitigating risks associated with model changes or deprecation.

Customers utilizing Forge have access to Mistral’s extensive library of open-weight AI models, which includes smaller yet powerful models such as the recently introduced Mistral Small 4. According to Timothée Lacroix, Mistral co-founder and chief technologist, Forge is instrumental in maximizing the value derived from these existing models. He noted that while smaller models inherently involve trade-offs in general topic proficiency compared to their larger counterparts, the ability to customize them empowers users to prioritize and emphasize specific knowledge areas. Although Mistral provides expert advice on optimal models and infrastructure, the final decisions remain with the customer.

Furthermore, for organizations requiring more than just guidance, Mistral Forge includes a dedicated team of forward-deployed engineers (FDEs). These FDEs embed directly within customer environments, assisting in identifying and surfacing the most relevant data, and adapting solutions to meet specific needs—a collaborative model inspired by successful practices at companies like IBM and Palantir. Salamanca added that Forge inherently provides all necessary tooling and infrastructure for generating synthetic data pipelines. She highlighted the value FDEs bring in helping enterprises, which often lack the specialized expertise, to build appropriate evaluations and ensure sufficient, high-quality data for their AI initiatives.

Mistral Forge has already garnered significant attention and is available to a diverse range of partners and early adopters. These include telecommunications giant Ericsson, the European Space Agency, Italian consulting firm Reply, and Singapore's DSO and HTX. Dutch chipmaker ASML, which notably led Mistral’s Series C funding round in September 2024 with an €11.7 billion valuation (approximately $13.8 billion), is also among the early partners. Marjorie Janiewicz, Mistral’s chief revenue officer, outlined the anticipated primary use cases for Forge, which encompass governments needing models tailored for specific languages and cultures, financial institutions with stringent compliance requirements, manufacturers requiring extensive customization, and technology companies aiming to fine-tune models to their unique code bases.