Cloudera Partners With Mistral to Run AI Without Moving Sensitive Data

Cloudera has struck a nine-figure partnership with Mistral AI to bring Mistral’s frontier models and Forge customization platform into Cloudera’s hybrid data and AI platform. The companies are pitching a route for enterprises to run and tailor AI around proprietary information without moving sensitive data outside their existing environments.

The partnership was announced at Cloudera’s EVOLVE26 event in São Paulo. Mistral’s models are to be integrated into Cloudera’s platform for deployment across public-cloud, private-server, hybrid, and edge environments.

Serving models and shaping them

Cloudera says customers will be able to run inference, generative AI, and agentic workflows alongside their data, rather than send it to another location or access models through an external API. The company frames that arrangement as important for regulated organizations weighing security, compliance, and data-privacy risks around sensitive information.

Forge adds the customization layer

Forge is the part of the deal aimed at changing model behavior, not merely hosting a model. Mistral introduced the system in March for enterprises building models grounded in proprietary knowledge. Cloudera says customers will be able to customize models with their own intellectual property while retaining control over data, infrastructure, and economics.

What Mistral says Forge can use

• Internal documentation, codebases, structured data, and operational records.
• Post-training methods to refine behavior for specific tasks and environments.
• Reinforcement learning to align models and agents with internal policies and objectives.

Control still needs testing

The deal separates two jobs: serving a model near the data and tailoring it to an organization’s own knowledge. Mistral says Forge includes evaluation frameworks for testing customized models against internal benchmarks, compliance rules, and domain-specific tasks before production deployment. That puts validation alongside control: keeping data in place addresses one enterprise constraint, while testing determines whether a tailored model fits the intended work.