Databricks evolves Delta Sharing into Linux Foundation-led OpenSharing
Databricks is evolving Delta Sharing into OpenSharing, an open protocol governed by the Linux Foundation for sharing data and AI assets across platforms. The expanded specification covers tables, unstructured data, agent skills and AI models while preserving compatibility with existing Delta Sharing deployments.
- OpenSharing is the next evolution of Delta Sharing.
- The protocol adds support for agent skills, AI models and Volumes.
- Apache Iceberg REST Catalog clients can access shared data.
- On-premises and private cloud data can be shared without migration.
- Existing Delta Sharing clients and integrations remain compatible.
From live tables to the broader AI stack
Delta Sharing was introduced at Data + AI Summit 2021 to let organizations share live data with partners, customers and other teams without exporting files, creating brittle ETL jobs or requiring recipients to use the same platform. Built on Delta Lake and Parquet, the open protocol allowed recipients to query shared tables in real time through tools including Apache Spark, Power BI, Tableau and pandas.
According to Databricks, Delta Sharing became the most widely adopted open protocol for secure data sharing, with more than 28,000 data recipients by the time OpenSharing was announced. Roughly a third of shares were flowing across platforms through open connectors rather than through Databricks-to-Databricks sharing, the company said.
Databricks says the needs of AI teams exposed the limits of a table-focused open protocol. Teams increasingly need to share agent skills, AI models and the semantic context used by agents, alongside the data those assets depend on. They also want to work across Apache Iceberg environments and share data held in on-premises or private cloud storage without migrating it first.
What OpenSharing changes
Databricks and the Linux Foundation jointly announced OpenSharing on June 10, 2026. It has two layers: an open source protocol and specification, and Databricks OpenSharing, the enterprise implementation integrated with Unity Catalog, Databricks Marketplace and Clean Rooms.
The specification, contributed by Databricks, is developed under the Linux Foundation AI & Data Foundation. It is licensed under Apache 2.0 and hosted on GitHub, where vendors and community members can implement it independently of Databricks. Open source Unity Catalog, also an Apache 2.0 project governed under the Linux Foundation, provides a vendor-neutral catalog and governance layer that can be paired with an open sharing server outside Databricks.
OpenSharing introduces a Share to Schema to Asset hierarchy. In addition to Tables, the specification includes Volumes, which are directories of files in any format, Agent Skills and AI models. This allows a provider to package tables and the AI assets built on them through one protocol.
The protocol also supports Apache Iceberg REST Catalog compatible clients, extending open sharing to Iceberg-compatible engines. Through the Databricks Storage Ecosystem, storage vendors can run an OpenSharing server connected directly to Unity Catalog, allowing data in on-premises or private cloud storage to be governed and shared without migration.
Compatibility and new sharing capabilities
OpenSharing keeps the existing Delta Sharing workflow. Current clients and integrations remain compatible, and existing setups continue to work. Databricks says organizations can enable the new asset types on a Unity Catalog metastore using the same share, recipient and grant process.
The zero-copy model also remains in place. Providers do not move data to share it. Instead, the sharing server issues temporary, scoped credentials, while the server stays outside the data path.
Databricks has introduced capabilities on top of OpenSharing, including SecureConnect, which is generally available and provides a Databricks-managed proxy that removes the need to configure firewall or network rules for every new recipient. Global Distribution is in Private Preview and replicates shared data closer to recipients across regions and clouds. Genie Agent Sharing is in Beta, while Lakebase table and change data feed sharing is in Public Preview.
OpenSharing is currently at the Linux Foundation AI & Data Foundation's sandbox stage. Databricks co-founder and CTO Matei Zaharia said, "OpenSharing extends that principle to the full AI stack, while expanding the cross-platform ecosystem to Iceberg recipients and on-premises providers." For users, the change provides a broader way to share governed data and AI assets without abandoning existing Delta Sharing connections. For the market, it places that interoperability under an open specification rather than limiting it to a single platform.
Databricks: From Delta Sharing to OpenSharing: what changed, what's next
