Apache Iceberg Tables
An Iceberg table stores its data in the open Apache Iceberg format, as Parquet data files tracked by Iceberg metadata files, either in your own cloud storage through an external volume or in Snowflake storage. Snowflake reads and writes it with near-native performance, including micro-partition-style pruning, rather than the slower external-table read path.
The distinction from a plain external table matters: external tables are read-only, query-time-only, with weaker pruning: they prune only on partition columns and have no per-column statistics like native tables. Iceberg tables, backed by Iceberg's own metadata and manifest files, let Snowflake both read and write them, and other engines (Spark, Trino, Athena) can read the same files, since Iceberg is an open table format. Who may write depends on the catalog: with Snowflake as the catalog, Snowflake writes, and other engines can too through Horizon Catalog; with an external catalog, the other engine writes, and Snowflake can too through an Iceberg REST catalog.
CREATE ICEBERG TABLE events_iceberg EXTERNAL_VOLUME = 'my_ ext_ volume' CATALOG = 'SNOWFLAKE'BASE_LOCATION = 'events/' AS SELECT order_id, order_date, amount FROM orders;
Iceberg tables exist for organizations standardizing on an open lakehouse architecture, where multiple compute engines beyond Snowflake need write access to the same physical data, without vendor lock-in to Snowflake's proprietary storage format.
The catalog choice decides what Snowflake may do with the table. With CATALOG = 'SNOWFLAKE', Snowflake owns the Iceberg metadata and the table gets full DML, ACID, and near-native performance; other engines can read and write it through Snowflake Horizon Catalog's Iceberg REST interface. Point the table at an external catalog instead and that catalog owns the table's state: another engine usually writes, and Snowflake reads what it committed. Through an Iceberg REST catalog integration (AWS Glue's REST endpoint, Unity Catalog, Open Catalog and others), Snowflake can also write to it with INSERT, UPDATE, DELETE and MERGE. The external volume in the example is the third piece, an account-level object holding the IAM relationship to the customer's bucket, playing the same role for Iceberg that a storage integration plays for stages; BASE_LOCATION is the table's folder inside it. Snowflake-managed catalog when Snowflake is the primary engine, external catalog when it's a guest at someone else's lakehouse.
SNOWFLAKE-MANAGED · CATALOG = 'SNOWFLAKE'
COMPUTE
ENGINESnowflakereads and full DML
commits
CATALOG
CATALOGSnowflake
reads and writes files through the external volume
YOUR CLOUD BUCKET
STORAGEData filesParquet + Iceberg metadata files
EXTERNALLY MANAGED · CATALOG INTEGRATION
OTHER ENGINESpark, Trino…writes the table
commits
CATALOGAWS Glue, Open Catalog or any REST catalog
ENGINESnowflakereads; writes only via REST catalog (Oct 2025)
reads files through the external volume
STORAGEData filesParquet + Iceberg metadata files
CATALOG = 'SNOWFLAKE' Snowflake commits and has full DML. Bottom: an external catalog owns the table's state; Snowflake reads it, and since October 2025 can also write through an Iceberg REST catalog integration. On an external volume, your cloud provider bills storage and there's no Fail-safe. Snowflake-managed tables can instead use Snowflake storage, billed by Snowflake, with Fail-safe on permanent tables.
