Databricks Geos: AI data residency

This article describes how Databricks Geos manage data residency when processing customer content for some AI capabilities, known as Designated Services.

How does Databricks manage data residency when processing customer content?

In traditional (both Classic and Serverless) Databricks architecture, there is a “compute plane” and a “control plane” that are provided on a region-specific basis.

Designated Services, including those that power Generative AI workloads, are available on a Geo basis. Designated Services are always managed by Databricks on the customer’s behalf.

Customer content is only processed in the same Geo as your workspace.

What is a Databricks Geo?

A Geo is a group of data center regions that Databricks groups together to give customers predictability and transparency regarding where your data is processed.

For Designated Services that use Geos, Databricks only processes data within the same Geo as your workspace.

The following table describes the available Databricks Geos and the regions it includes:

Databricks Geo

Comprising regions in:

Americas

  • United States

  • Brazil

  • Canada

Asia

  • Hong Kong

  • Japan

  • Korea

  • Singapore

Australia and New Zealand

  • Australia

  • New Zealand

Europe

  • European Economic Area (EEA)

  • Switzerland

  • United Kingdom

India

  • India

Mainland China

  • Mainland China

Middle East and Africa

  • Qatar

  • South Africa

  • United Arab Emirates

Designated Services offered on a Geo basis

For Designated Services, such as those that depend on GPUs or other advanced processing, customer content is processed in the workspace Geo. For more information see Databricks Designated Services

Will my data be sent out of Geo?

Customer content is processed in Geos similar to how classic Databricks architecture processed data within a specific region.

Customer content is only processed within the same Geo as a customer Workspace.

Will my existing compute planes be moved to a different region in the same Geo?

No. In both serverless and classic Databricks architectures, regional compute planes remain in their region for all existing services. In the classic Databricks architecture, the compute plane operates in an account under your control, and you must specify the region in which it operates. Serverless components offered on a regional basis, such as Serverless SQL warehouses, do not process data outside of your selected region.

How do Geos affect Preview features?

Preview features may involve processing data across Geos unless otherwise noted in the documentation.

Additional resources