Cloudera and Mistral AI have formed a partnership aimed at letting enterprises build and run customized AI systems against governed data while keeping data, models and compute inside customer-controlled environments.

The companies are targeting organizations in regulated and data-intensive sectors that want to use generative and agentic AI without moving sensitive information into a public external service.

Mistral said the collaboration will combine its open-weight models and customization capabilities with Cloudera’s hybrid data and AI platform, which operates across public cloud, private infrastructure and on-premises environments.

Partnership centers on control of data and models

The companies describe the approach as sovereign AI, meaning customers can keep data within defined boundaries, adapt and own models based on open weights, choose where training and inference run, and govern the systems without giving an external platform control of the full learning loop.

That can matter in financial services, manufacturing, telecommunications and other sectors where data residency, intellectual property or operational requirements limit which information can leave a controlled environment.

Cloudera says its platform manages about 30 exabytes of customer data. Mistral is positioning its models as a way to bring AI closer to that data rather than requiring organizations to move the underlying information into a separate AI service.

The partnership also gives customers a route to customize models around domain-specific workflows instead of relying only on general-purpose models. Mistral’s model-development and customization tooling can be used alongside Cloudera’s governed data environment so training and inference remain tied to the customer’s infrastructure choices.

Sovereignty is becoming an enterprise architecture question

The term “sovereign AI” is used broadly in the industry and can refer to national infrastructure, local data residency, customer ownership of models or some combination of those ideas.

In the Cloudera-Mistral partnership, the emphasis is operational control: customers choose the infrastructure and jurisdiction, retain control over their data and can run adapted models without sending sensitive information to a shared external endpoint.

That does not remove the need for governance. Organizations still have to decide which datasets a model can access, how training data is managed, which users and agents have permissions, and how outputs and actions are monitored.

TNGlobal recently reported that a WD-sponsored IDC survey found AI adoption is increasing enterprise data retention and storage demand. As organizations keep more data and reactivate older information for AI workloads, the question of where that data is stored and where models are allowed to process it becomes more important.

Partnership targets hybrid and air-gapped deployments

Cloudera and Mistral say the combined approach can support deployments across cloud, on-premises and edge environments, including sovereign and air-gapped infrastructure.

Air-gapped environments are physically or logically isolated from public networks and are used for some highly sensitive workloads. Supporting AI in those settings requires more of the model-serving, data, governance and monitoring stack to run locally.

The companies did not disclose financial terms for the partnership or provide customer deployments tied specifically to the new arrangement.

The announcement instead reflects a broader enterprise AI trend: as organizations move beyond pilots, model choice is increasingly being considered alongside data location, governance, infrastructure cost and control. Cloudera and Mistral are betting that some large enterprises will prefer to bring models to their data rather than bring sensitive data to a general-purpose AI service.

WD-sponsored IDC survey finds AI is extending enterprise data retention and storage demand