Forge Use Cases

Four Production Deployments

Mistral Forge is past the demo stage: leading organizations are using it to transform their operations. We’ll walk through four deployments across four high-stakes sectors—semiconductors, space observation, telecommunications, and defense. Examining them one by one reveals the boundary between what a general-purpose model can do and what it will never do without your proprietary knowledge.

ASML: Cutting Hours Off Failure Diagnostics

ASML is the global leader in lithography machines used to manufacture the world’s most advanced chips. These machines run 24/7 in customer factories, and the speed and precision they demand rely on extremely sophisticated models. Two critical challenges drove the partnership with Mistral: failure diagnostics, since every hour of downtime represents massive losses for chip manufacturers and maintenance engineers must identify the root cause ASAP; and code quality, since in ASML’s development cycle, a bug caught early is a bug that never reaches the customer.

This led to two custom solutions. The diagnostic agent combines the expertise of seasoned ASML engineers with a custom Mistral model via Forge: it analyzes machine data from a new angle and proves 120 times faster than previous methods at comparable accuracy. The code review agent takes the form of custom AI agents that adhere to ASML’s security and data protection requirements; they act as permanent reviewers, flag risky changes, and raise overall software quality. The benefit isn’t just speed: defect detection shifts upstream, before customer delivery, delivering measurable efficiency gains across the entire value chain.

European Space Agency: Mining Satellite Archives

The ESA manages a massive volume of Earth observation data collected by its Copernicus satellites, which is essential for environmental monitoring, weather forecasting, and natural disaster management. The challenge lies less in availability than in utilization: the data is voluminous and complex, requiring specialized expertise, and scientists spend considerable time searching, filtering, and analyzing the records relevant to their research. The model developed with Mistral understands the technical vocabulary of Earth observation, facilitates searching satellite archives, assists in interpreting geospatial data, and automates part of the image classification and analysis process. Forge therefore applies to highly specialized scientific domains whose vocabulary has no equivalent elsewhere.

Ericsson: Operating 5G Networks at Global Scale

Ericsson is a global leader in telecommunications infrastructure, with networks deployed in over 180 countries, generating vast volumes of technical data. Optimizing a 5G network requires constant performance analysis, rapid anomaly detection, and forecasting capabilities to anticipate congestion and failures. Specialized models built with Forge analyze network logs to detect anomalies, automatically optimize parameters, forecast load for capacity planning, and assist engineers in diagnosing and resolving issues. Their advantage comes down to a simple point: they integrate Ericsson’s telecommunications terminology and proprietary protocols, whereas a general-purpose model has to guess.

DSO National Laboratories: Sovereignty as a Starting Constraint

DSO National Laboratories is Singapore’s defense research organization, where data sovereignty and security are top priorities. The targeted applications require models capable of operating in disconnected (air-gapped) environments, processing classified data without leakage risk, adapting to specific operational scenarios, and offering maximum reliability in critical contexts. The partnership with Mistral enables DSO to develop models deployed entirely on-premise, with no external connection: the model is delivered as a standalone artifact, no data leaves DSO’s infrastructure, and Forward-Deployed Engineers work on-site with the teams.

The Common Thread

These four deployments converge on three points. First, customization creates a competitive advantage: a model that integrates an organization’s institutional knowledge becomes a strategic asset that competitors cannot replicate with the same base model. Second, deployment flexibility matters as much as performance—ASML needs speed, the ESA needs scientific capabilities, Ericsson needs network integration, and DSO needs total isolation; only the full range of cloud, VPC, and on-premise options covers this diversity. Finally, human support has been decisive in every project: customizing a model is an engineering effort conducted jointly with your business teams, never a self-service task.

Key Takeaways

• ASML achieves failure diagnostics 120 times faster thanks to Forge
• The ESA leverages its Earth observation data with specialized AI
• Ericsson optimizes its 5G networks with models trained on proprietary data
• DSO deploys sovereign models in air-gapped environments for defense
• Customization creates a competitive advantage that competitors cannot replicate