What OpenSharing Is and Why It Matters
OpenSharing is an open AI model sharing protocol that standardizes how organizations publish and consume models, agent skills, and data assets across platforms and clouds, enabling secure, interoperable and vendor-neutral collaboration. Built as the successor to Delta Sharing and now a Linux Foundation project, OpenSharing extends open data sharing into the full AI stack, including unstructured data. Before OpenSharing, enterprises had to wire together custom integrations or commit to single-vendor marketplaces to exchange AI capabilities. Databricks claims Delta Sharing “proved the industry would choose open over locked-in,” and OpenSharing aims to repeat that outcome for agentic AI workflows. By defining common APIs for discovery, authorization, and access, the protocol treats AI assets like reusable components rather than siloed products, addressing a central pain point in enterprise AI interoperability and cross-platform data collaboration.
From Delta Sharing to OpenSharing in the Agentic Era
Databricks introduced Delta Sharing in 2021 as an open standard for secure data collaboration, and it has since become, according to the company, “the most widely adopted open data-sharing protocol.” OpenSharing is the next chapter, expanding from tabular data into agent skills, AI models, and unstructured content. The protocol keeps the zero-copy sharing philosophy: asset providers publish once, while recipients across many platforms gain direct access without duplicating data. New support for Apache Iceberg APIs means providers can reach Iceberg-native tools through the same interface, widening the ecosystem for cross-platform data collaboration. This strategy reflects Databricks’ move from a single platform to an infrastructure layer for data and AI, designed to sit underneath tools such as Spark, BI dashboards, or external AI services rather than pull customers into a closed environment.
Reducing Vendor Lock-In and Enabling Unified Governance
Enterprises adopting AI at scale face a familiar problem: every platform, cloud, and vendor prefers its own formats, storage layers, and marketplaces, creating tight coupling and vendor lock-in. OpenSharing targets this by giving providers and consumers a single, open protocol for AI assets. Structured data, models, and agent skills can be shared using common APIs, which helps teams combine best-of-breed tools without abandoning unified governance. Data providers can distribute proprietary agent skills that run directly on their source systems, instead of shipping files to be copied and maintained everywhere. LSEG’s decision to use OpenSharing for its “LSEG Everywhere” strategy highlights this appeal; as the company states, it “allows any customer to use our data in any tool or any cloud with any model.” Governance teams retain control at the source, while application teams gain flexibility.
Bringing Cloud AI to On-Premises and Multicloud Environments
Many organizations still keep sensitive or regulated data on-premises or in private clouds, yet they want the latest AI capabilities often associated with public cloud platforms. OpenSharing addresses this by integrating with on-premises storage partners such as Everpure, MinIO, and Qumulo, with more providers including HPE, NetApp, and others expected. These integrations allow cloud AI platforms to connect directly to on-premises data with no physical data movement, aligning with enterprise requirements for security and compliance. Enterprises can run modern analytics and AI agents over data that never leaves their controlled environments, while sharing specific models or skills outward when needed. This design strengthens OpenSharing’s role as an interoperability layer across cloud and on-premises systems, supporting multicloud and hybrid architectures without forcing organizations into a single infrastructure vendor.
OpenSharing’s Role in Databricks’ Broader AI Strategy
OpenSharing arrives alongside Databricks’ broader push around the Data + AI Summit, where more than 30,000 professionals are expected to gather to see new data and AI product releases. Summit programming spans 800+ breakout sessions and 25+ hands-on courses, with themes including agentic architectures and real-world AI applications, indicating how central cross-platform data collaboration and AI interoperability have become. Keynotes from Databricks founders and guests such as leaders from Microsoft and OpenAI frame Databricks not as a closed suite but as a foundational data and AI layer. OpenSharing fits that positioning: rather than locking customers into a proprietary ecosystem, the protocol is released under the Linux Foundation and published on GitHub, signaling that Databricks wants to anchor an open, ecosystem-driven approach to enterprise AI and vendor lock-in avoidance.
