For about a decade, the standard advice from the technology industry was simple: move your data to the cloud, and then modern AI and analytics tools will work on it. The problem is that a growing number of large companies cannot do that, and many never will.
Regulators tell banks and pharmaceutical companies where their data must stay. Classified engineering data at semiconductor firms is legally restricted to on-premises servers. Trading firms carry decades of financial records where the sheer cost of transferring the data to a cloud service makes migration economically absurd. These are not old-fashioned companies resisting change. They are sophisticated organizations with real constraints.
Databricks, a major platform used by companies to run AI and data analysis on corporate data, is directly addressing this with today's launch. The program introduces a shared technical standard called OpenSharing, which the Linux Foundation also announced today as an independent, open project. The idea is that storage vendors install a small connection layer using this standard on their systems. Once that is in place, Databricks can read and analyze the data where it sits: in your data center, at a remote site, or inside a private network. Nothing moves.
Four storage vendors are live at launch: MinIO, Everpure, Qumulo, and VAST Data. Six more, including HPE, NetApp, Nutanix, Cohesity, Commvault, and Rubrik, have committed to follow by end of year. Collectively, these providers manage a very large share of the world's corporate on-premises data.
The cost argument is not subtle. Cloud providers charge per gigabyte when data leaves their network, a fee called egress. Moving 50 terabytes out of a cloud service to another provider or back on-premises can cost between $3,500 and $7,000 in transfer fees alone, before any migration work begins. For organizations measured in petabytes, the economics simply do not work.
Beyond cost, there is a compliance dimension that is growing, not shrinking. Around 75% of large enterprises now cite data sovereignty, meaning legal requirements about where data can be stored or processed, as a driver of their infrastructure decisions. In Europe alone, sovereign cloud deployments grew 32% in a single year.
Gartner named hybrid computing, meaning systems that work across on-premises servers and cloud services simultaneously, as the top infrastructure trend for 2026. That shift is already visible: 87% of enterprises now operate across hybrid or mixed environments.
For a business operator, the practical read is this. If your company uses or is evaluating Databricks and you have data you have not been able to connect to it because it lives on your own servers, this launch directly removes that barrier. You talk to your storage vendor, they connect to this standard, and your existing Databricks setup can now access that data.
For anyone not yet using Databricks, the more important signal is directional. The industry assumption that AI requires cloud migration is breaking down. Major platforms are racing to meet data where it lives rather than demanding you bring data to them. That changes the calculation for every organization that has delayed AI programs because of data location constraints.
One honest note: querying data remotely over a network rather than from local cloud storage can introduce some speed trade-offs. For real-time, high-frequency operations, that matters. For analysis, reporting, and model training against historical records, it likely does not. The partner program currently covers structured data, meaning organized tables and databases. Support for unstructured data, documents, images, video files, is on the roadmap but not available yet.