Forge Partners

Real-world cases that prove Forge’s value

Mistral Forge has moved well beyond the concept stage. At GTC in March 2026, Mistral AI announced several inaugural partnerships with leading organizations. These four public cases matter less for the specific organizations involved and more for what they reveal about the conditions under which a Forge project delivers value. Read them looking for similarities to your own context.

ASML: Turning technical documentation into an expert assistant

ASML is the global leader in lithography systems for semiconductor manufacturing. This Dutch company operates in a highly specialized market with extremely dense technical documentation and proprietary vocabulary that generalist models simply don’t know. Ask a standard model about an ASML component, and you’ll get a plausible but useless answer because it has never seen those specifications.

Forge creates a specialized model here that analyzes and synthesizes thousands of pages of internal specifications, assists engineers in diagnosing complex machine issues, and accelerates onboarding for new hires by giving them an assistant that truly knows ASML systems. The lesson translates directly: whenever an organization owns voluminous, specialized technical documentation, Forge converts that mass into accessible expertise for everyone. You measure ROI in saved research time and shorter onboarding cycles.

ESA: Capturing knowledge across multi-year projects

The European Space Agency (ESA) manages projects of extreme complexity, with development cycles spanning several years and documentation scattered across dozens of industrial partners. Tracing technical decisions here isn’t just best practice—it’s a regulatory requirement, which rules out any approximate answers.

ESA uses Forge to search and cross-reference information across massive corpora—mission reports, technical specifications, lessons learned—assist engineers in drafting documents that comply with ESA standards, and capture knowledge from past missions for future projects. That last point will resonate with any organization managing complex, multi-partner projects: consulting firms, engineering offices, government agencies. The symptoms are the same—scattered information, expertise concentrated in a few people, risk of knowledge loss when staff leave.

Ericsson: Integrating a fast-moving regulatory corpus

Ericsson is a global leader in telecommunications infrastructure, a sector characterized by extremely dense technical standards—3GPP, 5G, Open RAN—and a rapid pace of change. These two properties combined explain why Ericsson chose Forge over simple document search.

The custom model masters 3GPP standards and 5G specifications, assists R&D teams in analyzing new standards and their product impact, and automates generation of technical documentation that complies with industry standards. When an engineer receives a new specification version, they rely on the model to identify affected points instead of manually re-reading the entire corpus. This logic applies to any heavily regulated sector—telecom, finance, healthcare, energy—where documents must natively respect industry standards.

DSO National Laboratories: Forge in a sovereign environment

DSO National Laboratories is Singapore’s defense research organization. Confidentiality and security requirements here are maximal, and data cannot leave national infrastructure, which immediately rules out any architecture relying on remote services.

The deployment is therefore entirely on-premise, with no connection to Mistral servers, using a model trained on classified data and total control over infrastructure, access, and results. This partnership serves as a proof point: if Forge works in this setting, it works in yours. Organizations subject to absolute sovereignty requirements—defense, intelligence, critical infrastructure—now have a genuinely viable on-premise option, not a degraded compromise.

What these four cases have in common

The first common thread is a massive, specialized document corpus: thousands, even millions of pages of technical documentation, where the combination of volume and specialization makes generalist models insufficient. The second is a measurable productivity challenge: the time spent searching, analyzing, and synthesizing information represents a significant portion of expert work—a portion that Forge reduces in quantifiable ways. The third involves security and compliance requirements: intellectual property, classified data, industry standards, all addressed through encryption, isolation, and on-premise deployment. The fourth, most often overlooked, is the presence of a business sponsor: each project is led by an R&D, engineering, or operations directorate, never by IT alone, and this sponsor defines objectives, validates results, and measures ROI.

Test this against your own organization

Take five minutes and answer honestly. Do you have a document corpus exceeding 10,000 pages in a specialized domain? Have you found that a generalist model with prompt engineering isn’t enough for your critical tasks? Does the expected gain—time saved, errors avoided, productivity—justify a 3-to-6-month investment? Is a business sponsor ready to champion the project and report on results? Can your data team prepare and structure the training data? Four yes answers out of five indicate that Forge deserves a thorough evaluation with the Mistral team; below that, focus first on prompt engineering and RAG.

Key takeaways

• ASML, ESA, Ericsson, and DSO illustrate the diversity of Forge use cases—from high tech to defense
• The common thread is a massive, specialized corpus that generalist models cannot master
• On-premise deployment proves Forge works in the most constrained environments
• Evaluate Forge’s relevance for your organization using the five key questions