Mistral AI launches Forge platform to let enterprises build custom AI models from scratch

Mistral AI Targets Enterprise Gap with Build-Your-Own AI Platform

Mistral AI announced Mistral Forge on Tuesday at Nvidia's GTC conference, introducing a platform designed to address a critical problem in enterprise AI: most projects fail because models trained on generic internet data don't understand specific business contexts1

1

. The French startup is betting that letting companies build custom AI models using proprietary data will set it apart from rivals OpenAI and Anthropic in the intensifying competition for enterprise clients1

1

.

CEO Arthur Mensch says the company's laser focus on corporate customers is paying off, with Mistral AI on track to surpass $1 billion in annual recurring revenue this year1

1

. The launch represents a strategic shift from simply sharing AI models to powering comprehensive enterprise AI strategies.

Privacy-Focused AI That Trains Models From Scratch

What distinguishes Forge from competitors is its approach to customization. While several companies in the enterprise AI space offer similar capabilities through fine-tuning or retrieval-augmented generation (RAG), these methods don't fundamentally retrain models1

1

. Instead, they adapt or query existing models at runtime using company data5

5

.

Mistral Forge enables enterprises to train models from scratch, potentially addressing limitations in handling non-English or domain-specific data while providing greater control over model behavior1

1

. "What Forge does is it lets enterprises and governments customize AI models for their specific needs," Elisa Salamanca, Mistral's head of product, told TechCrunch1

1

.

The platform is positioned as a privacy-focused AI tool designed for industries where data privacy is critical, such as finance, defense, and manufacturing4

4

. Unlike rivals that rely on cloud services, Forge allows companies to build custom language models on their own systems using only their private data4

4

.

Building on Open-Weight Models and Expert Support

Forge customers can build their solutions using Mistral's library of open-weight models, including the recently introduced Mistral Small 41

1

. The new multimodal model features 119 billion parameters with 6 billion active ones per token, automatically switching between models depending on the task2

2

. This results in a 40 percent reduction in end-to-end completion time and three times more requests per second compared to Mistral Small 32

2

.

Co-founder and chief technologist Timothée Lacroix explained that smaller models can't match larger counterparts across every topic, but "the ability to customize them lets us pick what we emphasize and what we drop"1

1

. While Mistral advises on which models and infrastructure to use, final decisions remain with customers1

1

.

For teams needing more than guidance, Forge includes forward-deployed engineers who embed directly with customers—a model borrowed from IBM and Palantir1

1

. The platform comes equipped with tooling and infrastructure to generate synthetic data pipelines, though Salamanca noted that enterprises often lack expertise in building proper evaluations and determining adequate data volumes1

1

.

Early Adopters Signal Strategic Focus Areas

Mistral has already made Forge available to partners including ASML, Ericsson, the European Space Agency, Italian consulting company Reply, and Singapore's DSO and HTX1

1

. ASML, the Dutch chipmaker, led Mistral's Series C funding round last September at an €11.7 billion valuation, approximately $13.8 billion at the time1

1

.

According to Mistral's chief revenue officer Marjorie Janiewicz, expected use cases include governments needing models tailored for their language and culture, financial players with high compliance requirements, manufacturers with customization needs, and tech companies requiring models tuned to their code base1

1

.

The timing aligns with Nvidia's push for enterprises to adopt its powerful new GPUs for on-premises training4

4

. Mistral positions itself as the top choice for companies wanting customizable AI that runs on Nvidia hardware without relying on US cloud giants4

4

. The French company raised over $2 billion in Series C funding in September 2025, making it the most valuable European AI company at a $14 billion valuation4

4

.