The Linux Foundation, together with Databricks, announces OpenSharing. This is an open protocol for the secure exchange of AI models, agent skills, and data between organizations and platforms, in the cloud and on-premises.
Databricks and the Linux Foundation introduce OpenSharing. This neutral protocol aims to make it possible to share and link data and AI assets without departing from zero-copy principles. The project is managed by the Linux Foundation and builds on Delta Sharing. The announcement comes ahead of Databricks’ Data + AI Summit in San Francisco.
OpenSharing can be seen as the next step after Delta Sharing, which Databricks launched in 2021 as part of the open Delta Lake project. While Delta Sharing focused on the secure sharing of structured data, OpenSharing expands that to the entire AI stack.
Sharing without copying
The core problem OpenSharing aims to solve is the lack of an open standard for exchanging AI components. Without such a standard, organizations resort to custom integrations or closed marketplaces from a single vendor. OpenSharing provides standard APIs for discovering, authorizing, and accessing agent skills and AI models, regardless of the platform.
The protocol also expands the existing Delta Sharing ecosystem with support for Apache Iceberg clients. This allows data providers to reach a larger number of recipients via a single protocol, including those working with Iceberg-native tools.
Equally important are the capabilities OpenSharing brings to hybrid environments. Through the protocol, organizations that keep data on-premises or in a private cloud can connect directly to cloud platforms and modern AI tools running there, without the need to move data.
Everpure, MinIO, and Qumulo already support OpenSharing as a managed service. Big names including HPE, NetApp, Nutanix, as well as Rubrik, Commvault, and Cohesity will follow.
Companies should not have to choose between keeping sensitive data on-site and using modern AI and analytics platforms
AB Periasamy, co-CEO MinIO
“Companies should not have to choose between keeping sensitive data on-site and using modern AI and analytics platforms to derive value from that data,” says AB Periasamy, co-founder and co-CEO of MinIO. “OpenSharing opens up access to data that cannot be moved. This lays the foundation for unlocking the massive, untapped AI value hidden within enterprise data environments.”
Neutral standard
By housing the project within the Linux Foundation, Databricks explicitly aims to position OpenSharing as a vendor-neutral standard, similar to earlier open standards that shaped the internet and cloud infrastructure. The project is available on GitHub.
OpenSharing is an interesting and potentially important step in the development of AI infrastructure. Everyone now agrees that silos are detrimental within a modern environment that must support AI. Furthermore, the consensus seems to be that AI and data need to be close to each other.
Counterweight to the platform argument
There are several solutions for this. Platform players such as Snowflake or Salesforce are relatively open, allowing the connection of data sources. However, the gravity of an AI implementation lies within the platform itself, where the AI workloads run.
In theory, OpenSharing makes it possible to link AI capabilities to data in a less dependent manner. The openness of the standard allows for greater flexibility.
Openness is quite trendy in the context of data and AI. No company is under the illusion that it will play a key role in customers’ data and AI strategies in isolation. This works in favor of the new protocol: companies including Atlassian, LSEG, SAP, and Stripe indicate they are embracing OpenSharing because it allows them to share data and AI assets across any platform, any cloud, and any model.
The shared motivation is clear: an open foundation avoids vendor lock-in and lowers the barrier to collaboration in the AI era. In the run-up to its Data+AI summit next week, Databricks is showing that it wants to play a central yet open role in the development of the IT landscape.