Zero-copy sharing with Databricks, Snowflake and hyperscalers

Most large companies already have a data platform outside SAP: a lakehouse, a cloud warehouse or a hyperscaler analytics service. For years, getting SAP data into it meant extracting files or running replication tools and then rebuilding SAP's business meaning on the other side. BDC's answer is zero-copy sharing: publish a data product once and let other platforms read it without a second physical copy, where the partner integration supports it.

Two patterns: SAP-branded editions and BDC Connect

SAP offers two ways to work with partner platforms. First, SAP Databricks is a Databricks environment delivered inside BDC: data products are shared to it through Delta Sharing, data scientists work on them, and results can be shared back to Datasphere. SAP announced a similar SAP Snowflake edition in late 2025. Second, SAP BDC Connect connects BDC to a customer's existing platform for bidirectional sharing. At Sapphire in May 2026, SAP named Databricks, Snowflake, Google BigQuery and Microsoft Fabric as BDC Connect partners and announced BDC Connect for Amazon Athena with general availability planned for the second half of 2026.

What zero-copy means, and what it does not

A share is configured from the data product in the catalog: check that the data product and its APIs have the right release, lifecycle and functional status (SAP Learning names Active, Shareable and Current), choose Share, add the target system, and confirm. Administrators of the target platform then manage who can see it there. That last step matters: once data is visible in a partner platform, the partner's access model applies.

Contoso's case

Contoso's quality team runs a cloud lakehouse that holds field-failure data from fleet customers. Kenji's team wants to predict warranty claims for CMC-ALT-100 by combining that data with sales and production data from CMP. The old plan was a nightly file export. Noor proposes sharing the relevant SAP-managed data products through BDC Connect, joining them with the field data in the lakehouse, and sharing the prediction table back into Datasphere for a quality story in SAC.

Noor also writes down what the design does not change. Shared data products are still subject to Contoso’s data-protection rules: customer contact data shared into the lakehouse falls under the same retention and purpose limits as in CMP. Costs move too. Data that is not copied still has to be read, and partner platforms usually charge for the compute that reads it, while BDC consumes capacity units. Finally, the share depends on the data product’s lifecycle: when SAP releases a new version of an SAP-managed data product, the partner-side jobs that read it must be retested, exactly like an interface after a partner changes its message format.