Many enterprise AI initiatives fall short not due to a lack of technology, but because the models in use fail to grasp the specific context of the business. These models are typically trained on broad internet data rather than a company's own decades of internal documents, workflows, and unique institutional knowledge. This disconnect is precisely where French AI startup Mistral has identified a significant opportunity.

EMBED_PLACEHOLDER_0

On Tuesday, the company unveiled Mistral Forge, a new platform designed to empower enterprises to construct custom AI models trained exclusively on their proprietary data. The announcement was made at Nvidia GTC, Nvidia's annual technology conference, which has a strong focus on AI and agentic models for business this year. This launch represents a strategic move for Mistral, a company that has consistently focused on corporate clients while competitors like OpenAI and Anthropic have gained more traction with consumer products.

CEO Arthur Mensch asserts that this dedicated enterprise strategy is proving successful, with the company projected to exceed $1 billion in annual recurring revenue this year. A core part of this intensified enterprise focus involves granting companies greater control over their data and AI systems.

While several players in the enterprise AI market offer similar-sounding capabilities, most concentrate on fine-tuning pre-existing models or adding proprietary data layers through techniques like retrieval augmented generation (RAG). These methods do not involve fundamentally retraining the core model; they instead adapt or query it during operation. Mistral claims its approach is different, enabling companies to train models completely from the ground up.

Theoretically, this foundational training could overcome limitations of more common methods. Potential benefits include superior handling of non-English languages or highly specialized domain data, along with enhanced control over model behavior. It could also allow firms to train agentic systems using reinforcement learning and reduce dependence on external model providers, mitigating risks associated with third-party model changes or discontinuation.

Forge customers can develop their custom models utilizing Mistral's extensive library of open-weight AI models, which includes smaller options like the recently launched Mistral Small 4. According to Mistral co-founder and chief technologist Timothée Lacroix, Forge helps extract more value from these existing models. "The trade-offs we make when building smaller models mean they can't be as comprehensively capable as larger ones on every topic," Lacroix explained. "The ability to customize them lets us choose what to emphasize and what to downplay."

Mistral provides guidance on model and infrastructure selection, but the final decisions remain with the customer. For teams requiring more than advice, Forge includes access to Mistral's team of forward-deployed engineers. These specialists embed directly with customers to identify the right data and tailor solutions to their needs—a service model inspired by companies like IBM and Palantir.

"As a product, Forge comes equipped with all the necessary tooling and infrastructure to generate synthetic data pipelines," a company representative noted. "However, understanding how to build the right evaluations and ensuring you have sufficient, appropriate data is an area where enterprises often lack expertise. That's the critical value the forward-deployed engineers bring."

Mistral has already made Forge available to initial partners, which include Ericsson, the European Space Agency, Italian consulting firm Reply, and Singapore's DSO and HTX. Early adopters also feature ASML, the Dutch chipmaker that led Mistral's Series C funding round last September, valuing the company at €11.7 billion (approximately $13.8 billion at the time).

These partnerships highlight the primary use cases Mistral anticipates for Forge. According to Mistral's chief revenue officer Marjorie Janiewicz, key applications include governments needing to tailor models for specific languages and cultures, financial institutions with stringent compliance demands, manufacturers requiring customized solutions, and technology companies that need to fine-tune models for their proprietary codebases.