The Announcement
Databricks has launched OpenSharing, a vendor-neutral open protocol for sharing data and AI assets across platforms and organizations, now hosted under the Linux Foundation. The announcement represents a significant extension of Delta Sharing, the widely adopted open data-sharing protocol Databricks introduced in 2021, and expands it to cover agent skills, AI models, and unstructured data. OpenSharing also adds native support for Apache Iceberg API clients, broadening the recipient ecosystem beyond existing Delta Sharing consumers, and introduces on-premises connectivity through storage partners including MinIO, Qumulo, and Everpure, with additional partners including NetApp, HPE, Nutanix, and Rubrik coming soon. The breadth of ecosystem endorsement at launch, spanning OpenAI, SAP, Stripe, LSEG, Atlassian, and Amadeus, signals that this is not a speculative initiative but an actively deployed standard with real production commitments behind it.
Our Analysis
The Timing Is Not Accidental
This announcement lands at a moment when enterprise AI architecture is shifting from single-model deployments to multi-agent, multi-platform workflows. According to ECI Research’s 2025 AI Builder Summit survey, two-thirds of enterprise AI leaders have already implemented multi-agent collaboration, enabling agents to coordinate and delegate tasks, in live or pilot workflows. OpenSharing’s first-order design problem is exactly this scenario: agents and models need to consume data and skills from multiple sources, across organizational and platform boundaries, without the friction of custom integrations or the compliance risk of copying sensitive data.
Delta Sharing solved a narrower version of this problem for structured data. OpenSharing generalizes it. The inclusion of agent skills as a shareable asset class is the technically interesting move here. In a world where a travel company like Amadeus wants its proprietary agent skill to be consumable by any customer’s AI environment, a standardized discovery, authorization, and access API eliminates the bespoke integration work that currently makes that kind of capability sharing unscalable. That is a meaningful architectural contribution, not a product rebranding exercise.
What This Means for ITDMs
For IT decision-makers, the practical question is: does OpenSharing reduce the cost and complexity of building data and AI pipelines that cross organizational or platform lines?
The answer is yes, with some important qualifications. The zero-copy sharing model eliminates the data movement costs, security risks, and replication latency that come with extracting data to a neutral location. For regulated industries where data residency or sovereignty concerns make physical data movement difficult, the on-premises connectivity story is genuinely useful. Stripe, LSEG, and SAP have each made public statements framing OpenSharing as consistent with their existing multi-environment strategies. That is not vendor co-marketing language; those organizations have specific technical and compliance constraints that make locked-in data sharing protocols expensive to operate.
The Linux Foundation governance model also matters here. Enterprise buyers have learned to be skeptical of “open” standards that remain effectively controlled by a single vendor. Hosting OpenSharing at the Linux Foundation, alongside projects like Kubernetes, OpenSSF, and others, places it in an institutional framework that reduces the perceived lock-in risk. Whether the governance structure translates into genuine multi-vendor steering committee influence over time is worth monitoring, but the signal is directionally correct.
The financial case for ITDMs is straightforward. Custom data-sharing integrations are expensive to build and expensive to maintain. A standardized protocol that any partner can consume removes a class of integration work from the backlog. For organizations managing multi-cloud data estates, this is real budget relief.
What This Means for Developers
From a technical standpoint, the Iceberg API support is the headline for practitioners. The Delta versus Iceberg format debate has been a persistent source of friction in the data engineering community. By adding Iceberg IRC client support to OpenSharing, Databricks effectively sidesteps that argument: providers can publish once and reach consumers on either table format. That is pragmatic interoperability, and it removes a real integration headache for teams managing heterogeneous lakehouse environments.
The agent skill sharing model deserves closer attention from anyone building production agentic systems. The core problem it addresses is distribution and versioning of reusable agent capabilities across organizational boundaries. Think of it as a package registry, but for skills rather than libraries, with built-in authorization and zero-copy access semantics. The standard APIs for discovery, authorization, and access are what make this composable at scale. Without a shared protocol, every cross-organizational agent integration degrades into a custom API project with its own auth model, versioning scheme, and documentation gap.
The on-premises storage partner ecosystem is also technically significant. Cohesity, Commvault, NetApp, Nutanix, and Rubrik in the upcoming partner list covers a substantial portion of enterprise on-premises storage infrastructure. If those integrations deliver on the “no data movement” promise, development teams working in regulated environments gain access to cloud AI capabilities without the compliance gymnastics that typically accompany data extraction projects.
What’s Next
Adoption Will Be Tested by Governance Maturity
OpenSharing’s long-term success depends on whether enterprises can operationalize the protocol within their existing data governance frameworks. The zero-copy model is technically elegant, but it requires organizations to have clear, enforced policies on who can publish assets, who can receive them, and under what conditions access is revoked. That is a governance problem, not a technology problem, and many organizations are still working through it.
ECI Research has found that 62% of organizations say inconsistent cloud tagging and cost attribution across platforms is their most significant barrier to accurate forecasting. A similar attribution and ownership clarity problem will emerge with shared AI assets. If an enterprise publishes an agent skill via OpenSharing and that skill is consumed by a partner who then builds a revenue-generating product on top of it, the questions of ownership, update responsibility, and liability attribution need clear organizational answers before the technical plumbing matters.
The On-Premises Bridge Is the Near-Term Opportunity
The more immediate adoption curve is likely to be driven by the on-premises connectivity story rather than the AI asset sharing features. Enterprises with large on-premises data estates have been underserved by cloud AI platforms precisely because the data movement problem is hard and expensive. The upcoming additions of Cohesity, Commvault, HPE, NetApp, Nutanix, Rubrik, and VAST Data to the OpenSharing partner ecosystem will determine whether this becomes a genuine bridge for legacy data environments or remains a capability primarily useful for cloud-native organizations. Watch the partner integration quality closely in the second half of 2026. If those integrations deliver on the latency and security promises, the on-premises use case could drive faster enterprise adoption than the agent skill sharing features, simply because the pain is more acute and the budget impact is more immediate.
Stay Ahead of Application Development Trends
Get weekly analyst insights, research notes, event coverage, and AppDevANGLE updates delivered directly to your inbox.
Subscribe for Weekly Insights
Join technology leaders, practitioners, and GTM teams following the trends shaping modern software delivery.
Looking for deeper research access?
Explore ECI Research reports, survey insights, and market analysis through the ECI Research Portal.
