Amazon S3 Tables now automates the Hive-to-Iceberg migration
On November 4, 2026, Amazon S3 Tables added managed migration of Apache Hive tables to Apache Iceberg: free conversion, Snappy-to-Zstandard recompression, and continuous sync without interrupting running workloads. Teams still on Hive should inventory their Parquet tables in Glue and test the cutover now.
November 4, 2026. Amazon S3 Tables adds managed migration of Apache Hive tables to Apache Iceberg. November 4, 2026. The Hive-to-Iceberg conversion is free — you only pay for the data copied, and source objects in S3 Intelligent-Tiering keep their current access tier. November 5, 2026. The service is live in ten Regions, limited to Hive tables stored in Parquet and registered in AWS Glue. The whole thing fits in one sentence: AWS now absorbs the per-table migration pipeline that took weeks to build, and removes the last excuse for staying on Hive.
For a data platform still running on Hive, this is not one more announcement. It is the data lake modernization question finally posed without the cost that stalled it: Iceberg already promised ACID transactions, schema evolution, and time travel, but the cutover meant writing a migration pipeline for every table.
Iceberg delivers ACID, but the migration used to pay the bill
Apache Iceberg has become the reference table format for modern data lakes. It delivers what Hive lacked: ACID transactions, schema evolution without rewriting, time travel, and native integration with Spark, Flink, and Trino. Hive, by contrast, remains a metastore and a legacy format, where Parquet and ORC tables live without transactional guarantees and schema changes are surgery.
The blocker was never technical. It was operational: migrating a Hive table to Iceberg meant building a migration pipeline per table — extract, convert, validate, cut over — a project AWS itself prices in weeks. Across a lake of hundreds of tables, the return on investment was hard to defend against competing priorities. The result: entire teams stayed on Hive, not out of conviction but out of opportunity cost.
That is precisely the lock S3 Tables attacks. The managed migration validates, converts, and copies data while production workloads keep running against the source.
What the managed migration actually does
S3 Tables does not merely copy files. It orchestrates the cutover in steps the team can review one at a time.
- Validate, then convert. Hive tables registered in AWS Glue and stored in Parquet are read, converted to Iceberg, and written as S3 Tables.
- Continuous sync. Destination tables can be re-synced with source changes as often as needed, enabling a progressive cutover instead of a big-bang switch.
- Rollback. At any step the team can fall back to the source — a safety net hand-built pipelines rarely provide.
- Automatic recompression. Tables compressed with Snappy are recompressed to Zstandard (ZSTD) and compacted, cutting storage, reducing object count, and improving query performance.
The economics are unusual for a migration service: the Hive-to-Iceberg conversion is free, and you pay only for the data copied. Source objects in S3 Intelligent-Tiering keep their access tier, avoiding retrieval fees. In short, AWS is subsidizing the move to accelerate adoption of its table format.
One detail confirms the direction: a migration skill ships in the Agent Toolkit for AWS, letting AI agents orchestrate the migration workflow autonomously. The loop is now closed — an agent can inventory, convert, and validate hundreds of tables without human intervention, provided the governance around it gets written.
The scope: Parquet, Glue, and ten Regions
The service is in partial general availability, and the launch scope is specific — it deserves a careful read before celebration.
- Format: Hive tables in Parquet only, registered in AWS Glue, within the same account and same Region.
- Regions: US East (N. Virginia, Ohio), US West (Oregon), Asia Pacific (Mumbai, Singapore, Sydney, Tokyo), and Europe (Frankfurt, Ireland, Stockholm).
- Out of scope today: ORC tables, non-Glue metastores, and cross-account or cross-Region migrations.
For a platform that ticks those boxes — Parquet in Glue, single-account, single-Region — the cutover becomes a console or API action. For everyone else, the service points the direction without yet covering the need. The safe move is not to confuse service availability with coverage of your own data estate.
What it changes for a data platform
This is a strategy signal as much as a feature. AWS is pushing Iceberg as the default table format across Athena, Glue, EMR, and Redshift, with S3 Tables as its dedicated storage surface. Managed migration removes the adoption friction at the exact moment AI workloads and modern engines demand transactional guarantees Hive does not have.
For a data lead, the priorities write themselves.
- Inventory Hive tables in Parquet in Glue. That is the covered scope today, and the simplest to migrate.
- Test on a subset. Continuous sync and rollback let you validate the cutover without halting production.
- Measure the storage win. Snappy-to-ZSTD recompression and compaction shrink object counts — a direct saving on the S3 bill and on Athena latency.
- Do not stall on out-of-scope cases. ORC tables and multi-Region designs still need a homegrown pipeline or a wait for service expansion.
It is also worth measuring what the announcement does not say. The service manages the migration, not the governance: who migrates, under which permissions, with what cost control, remains the team’s job. The migration skill for AI agents is convenient, but an agent migrating hundreds of tables must be audited like any other code. Speed of cutover must not replace traceability of cutover.
Verdict
Amazon S3 Tables does not make Iceberg a better format — it already was. It removes the cost of the move, which was the real obstacle.
If your Hive tables are in Parquet in Glue, single-account and single-Region, migrate now: the conversion is free, continuous sync and rollback reduce risk to almost nothing, and ZSTD recompression pays for the move in storage alone. If you are on ORC, outside Glue, or multi-Region, lay the groundwork by cleaning up your Parquet tables and tracking scope expansion. The trap is not missing the announcement — it is staying on Hive out of inertia once AWS has taken the exit cost off the table.