Oracle adds Iceberg support to share company data across AI systems

Oracle wants company data to serve more AI and analytics systems without creating a separate copy for each one. Its October 28 announcement of Apache Iceberg support in AI Data Platform supplies a concrete mechanism: managed and external tables. That reuse depends on compatible engines and supported integrations—not simply choosing an open format.

Iceberg is an open table format: it defines how data files, table descriptions and changes are organized in data-lake storage. Oracle’s addition lets teams create new Iceberg tables inside AI Data Platform or register tables that already exist elsewhere. The distinction determines who manages the data and what happens when a table is removed.

Managed tables suit data created and maintained primarily in AI Data Platform. Teams can define columns and partitioning—the way data is divided for efficient access—using SQL, the language used to work with database tables, or Oracle’s interface. The platform manages both the table definition and its underlying data lifecycle.

External tables take a different route. Teams register an existing Iceberg table while its data stays in its established storage location. AI Data Platform then provides governed discovery, querying and transformation. Oracle positions this option for datasets maintained by another engine or shared among several consumers.

The ownership choice has a consequential deletion rule. Dropping a managed table removes both its definition and its data. Dropping an external table removes only its AI Data Platform definition; the underlying source data remains intact.

Oracle says an Iceberg data product can serve data engineering, analytics and AI workloads while remaining accessible to compatible services, including Autonomous Database through supported integrations. The intended benefit is reuse: teams can build a governed dataset once rather than maintain a separate copy for every environment.

The support accompanies Oracle’s October 26 Autonomous AI Lakehouse announcement, which included expanded native Iceberg functionality through external catalogs or Oracle AI Data Catalog. Oracle says users and AI agents can create, populate, update and delete Iceberg tables using Oracle SQL. Its product walkthrough also describes retaining existing supported catalogs, object storage and Iceberg-compatible engines.

Not every catalog capability is available now. Oracle’s announcement places the new AI Data Catalog capabilities within the next 12 months, including mounting distributed sources without duplication and automatically enriching their descriptions. That timetable is separate from the October 28 statement that AI Data Platform now supports managed and external Iceberg tables.

Inside AI Data Platform, both table types support inserts, updates, deletes, merges, schema changes and snapshot-based queries. Writes commit as complete changes, so readers see either the previous consistent table or the newly committed version—not a partially updated dataset. Pipelines can therefore apply new records and corrections without rebuilding the entire table.

Each committed change creates a snapshot. Historical queries can help reproduce a model-training dataset, investigate a data-quality problem or explain why an analytical result changed. Iceberg also separates a table’s logical structure from its physical files, allowing many column changes without routinely rewriting the table. Owners still need to assess downstream effects and validate changes.

The ongoing operating work

• Compact small files created by frequent writes; otherwise, scans can become less efficient as tables grow.
• Consolidate metadata files so query planning does not become slower as the table’s history expands.
• Monitor file counts, query latency and simultaneous workloads, then adjust partitioning and maintenance schedules to match actual use.