When Databricks pioneered Delta Sharing in 2021, we got down to remedy an issue that each information staff knew too nicely: sharing stay information throughout organizational boundaries was gradual, fragile, and stuffed with compromise. You both copied information — creating stale replicas and compliance complications — otherwise you constrained your self to solely sharing with companions on the identical platform as you, thereby considerably proscribing innovation.
Delta Sharing modified that. A single open protocol. No information copying. No platform silos. And within the 5 years since, it has change into probably the most extensively adopted open zero-copy data-sharing protocol — with 28,000+ information recipients and 33% of shares flowing throughout platforms by way of open connectors. Main firms reminiscent of SAP, Atlassian, Mercedes-Benz, The Commerce Desk, LSEG, S&P World, and lots of extra have adopted Delta Sharing to share and collaborate on information.
However the world has moved on. The rise of agentic AI has essentially modified what enterprises have to share. At the moment, we’re taking the following step.
We’re excited to announce OpenSharing — the following evolution of Delta Sharing, and the business’s first open protocol constructed for the agentic period. OpenSharing advances Delta Sharing into an unbiased open-source mission, increasing its scope from information sharing to the total AI stack: fashions, brokers — throughout any cloud, any vendor, and any format.
Why sharing protocols have to evolve for AI
Delta Sharing was constructed for a world of tables and recordsdata. However organizations should now trade semantic context, AI abilities, unstructured information, and autonomous brokers throughout cloud, vendor, and firm boundaries. At the moment’s sharing protocols stay locked into vendor-specific codecs, cannot deal with AI logic, and rely on brittle networking that takes weeks to configure for every new associate.
The outcome: collaboration slows, information silos persist, and the worth locked inside enterprise information goes unrealized.
OpenSharing solves this. It is a single open protocol that shares information and AI throughout any format, any cloud, and any organizational boundary — natively supporting Delta Lake, Apache Iceberg, and Parquet so information stays the place it lives and flows to whoever wants it.
“Delta Sharing proved the business would select open over locked-in. OpenSharing extends that precept to the total AI stack, whereas increasing the cross-platform ecosystem to Iceberg recipients and on-premises suppliers. The agentic period deserves an open basis, and OpenSharing delivers it.” — Matei Zaharia, Co-founder and CTO of Databricks.
OpenSharing on Databricks
OpenSharing exists at two layers. The open-source protocol — now hosted by the Linux Basis — is the printed spec that any vendor or neighborhood member can implement. Databricks OpenSharing is the enterprise implementation of the open protocol, constructed on high of different Databricks options reminiscent of Unity Catalog for governance and audit logging, Market for discoverability, and extra.
We’re excited to launch a set of options for OpenSharing on Databricks.
Genie Agent Sharing: share a ruled AI expertise, not simply information
For the primary time, organizations can share ruled AI experiences — not simply datasets — throughout organizational boundaries.
Genie Brokers are Databricks’ AI-powered conversational analytics environments. With OpenSharing, a supplier can now share Genie Brokers — together with their underlying semantic context, enterprise metrics, and reusable AI logic — with any associate or buyer, with governance end-to-end by way of Unity Catalog. Optionally, suppliers can management how recipients entry information — together with hiding proprietary Genie directions, proscribing information entry to the Genie Agent solely, setting each day immediate quotas, and capping row export limits. These controls unlock new monetization alternatives for information suppliers, reminiscent of usage-based pricing as a substitute of a full information license.
SecureConnect and World Distribution: easier multi-cloud networking, decrease egress prices
Cross-cloud information sharing has at all times had two distinct issues. OpenSharing on Databricks now solves each.
The primary is networking. When supplier storage sits behind a non-public community — which is nearly at all times the case for delicate information exchanges or regulated industries — onboarding a brand new recipient can take weeks of handbook IP allowlisting, firewall coordination, and back-and-forth with cloud admins. For suppliers with dozens or lots of of recipients, this does not scale. SecureConnect solves this drawback: a Databricks-managed proxy that routes storage entry on behalf of all recipients. Configure it as soon as — no per-recipient firewall modifications wanted, ever. Learn the announcement weblog.
The second is egress value. Cross-cloud queries generate egress charges that compound at scale, changing into a major and unpredictable value that makes broad multi-cloud sharing economically impractical. World Distribution solves this with computerized cross-region and cross-cloud replication. Recipients question a neighborhood reproduction — quick, with no egress charges. Suppliers get a predictable value construction. World groups get low-latency entry no matter the place the supply information lives.
Open Consumer Interoperability & On-prem Storage Ecosystem: Meet your companions the place they’re
OpenSharing is constructed on the conviction that information ecosystems thrive when they’re actually open — not simply in identify, however in observe. Which means supporting the codecs, storage techniques, and shoppers your companions already use.
Storage Ecosystem: Govern all the things, wherever it lives
Not all enterprise information can — or ought to — transfer to the cloud. Regulatory mandates, information gravity, edge latency, and sheer economics imply that a few of the world’s most dear information will keep on-premises. OpenSharing reaches it.The Databricks Storage Ecosystem brings the Databricks Knowledge Intelligence Platform on to on-premises, non-public cloud, and edge environments — powered by OpenSharing. Storage companions implement the OpenSharing server, connecting their information estates to Unity Catalog with out transferring a single byte. No migration. No duplication. Learn the announcement. Launch companions embody MinIO (GA), Everpure (Personal Preview), Qumulo (Personal Preview Quickly), and VAST Knowledge (Personal Preview Quickly) — with Cohesity, Commvault, NetApp, and Nutanix coming by the tip of the yr. Collectively, these companions handle lots of of exabytes of enterprise information.
Iceberg interoperabilityDelta Sharing is already supported in a variety of platforms and connectors, together with Databricks, Tableau, Energy BI, Apache Spark, and Snowflake. OpenSharing has now added help for Apache Iceberg REST Catalog API — making it potential to share information with any Iceberg-compatible consumer. Suppliers may share tables from exterior catalogs together with AWS Glue, Hive Metastore, and Snowflake Horizon — bringing exterior information into the ruled OpenSharing ecosystem with out replication.
How OpenSharing works
Constructing on the identical simplicity that made Delta Sharing profitable, OpenSharing extends the protocol to help the total AI asset stack:
The information supplier creates a share in Unity Catalog — defining which datasets, fashions, brokers, or Genie Brokers to share and setting fine-grained entry permissions.The recipient receives safe credentials and queries the share immediately from their current instruments, cloud, or Iceberg consumer — without having to be on Databricks.Unity Catalog enforces governance end-to-end — auditing each entry, imposing row- and column-level controls, and guaranteeing compliance insurance policies journey with each shared asset.Knowledge by no means strikes — recipients question stay information immediately from the supplier’s cloud storage, guaranteeing a single supply of fact.
For enterprise deployments on Databricks, SecureConnect and World Distribution layer on high of this move — dealing with cross-cloud networking and replication mechanically, with no modifications to how suppliers or recipients work together with their shares.
