Snowflake made Open Data Sharing generally available for live Apache Iceberg tables to any Iceberg REST Catalog client.
I read this as Snowflake admitting that customers will not move compute to get data. Providers can now serve supported Iceberg and dynamic Iceberg tables to Apache Spark and other IRC readers with no scheduled export and no copy to track, the company said.
Reader access is granted and revoked in Snowflake, powered by Apache Polaris. That control point matters because the protocol alone does not create a sharing relationship. Snowflake frames this as the third step after open table formats in October 2025 and catalog-linked databases for Polaris, Unity Catalog and AWS Glue in August 2026, followed by Open Data Sharing in October 2026.
The comparison that counts is distribution friction. Snowflake positions Open Data Sharing as a reproducible alternative to FTP exports or legacy Delta Sharing, recently rebranded as OpenSharing, without its recipient-activation and credential-file process. Against its own Marketplace, the contrast is also clear. Flexera notes that paid Marketplace listings and private sharing require an account configured for payments, which keeps that path inside Snowflake billing and identity.
The concession is scope. Only supported Iceberg tables and dynamic Iceberg tables can be shared directly, so other Snowflake native format tables must first be materialized into a dynamic Iceberg table. That extra step survives my claim rather than breaks it, because filtered, customer-specific tables are exactly what data products need. If you sell data, build for Spark first and treat Snowflake as the control plane, not the destination.





