Is Iceberg replacing Delta Lake?

Summary

• The public record shows two active projects. Delta Lake released 4.4.0 in August 2026 and 4.4.1 in October 2026, and Apache Iceberg released 1.12.0 in September 2026.
• Both are open source under the Apache License 2.0, with different homes: Delta Lake is an independent project within LF Projects under the Linux Foundation, and Iceberg is an Apache Software Foundation project.
• Their feature sets now overlap: deletion vectors, a variant type and row tracking (Iceberg calls it row lineage) appear both in Iceberg's version 3 spec and among Delta Lake's table features.
• Interoperability lets one copy of the data serve both formats. Delta Lake's UniForm generates Iceberg metadata so Iceberg clients can read Delta tables, and Iceberg documents an action for snapshotting supported Delta tables into Iceberg.
• On Databricks, Unity Catalog managed tables support both Delta Lake and Apache Iceberg, Delta Lake stays the default format, and Iceberg reads (UniForm) open Delta tables to Iceberg clients.

Is Iceberg replacing Delta Lake?

The projects' own release history and documentation don't show that. Delta Lake and Apache Iceberg both shipped releases in the past two months, both are developed in the open under the Apache License 2.0, and Iceberg's newest format version adds capabilities Delta Lake already documents. Both projects' docs describe ways to work from one copy of the data across formats, such as UniForm for Iceberg clients reading Delta tables and an Iceberg action that snapshots Delta tables. For a team choosing today, the useful question is which format fits the engines and catalog it runs.

What the release record shows

Delta Lake 4.4.0, released in August 2026, added Apache Spark 4.2 support, identity columns in Spark SQL DDL and generated columns for Delta tables created through Unity Catalog. Version 4.4.1 arrived on October 5, 2026. Apache Iceberg 1.12.0 shipped in September 2026, and the Iceberg spec now lists versions 1, 2 and 3 as complete and adopted by the community, with version 4 under active development.
Version 3 of the spec adds binary deletion vectors, row lineage tracking, default column values, multi-argument transforms for partitioning and sorting, table encryption keys and new types, including variant and geometry. Delta Lake's protocol documentation lists deletion vectors, row tracking, type widening and a variant type among its table features. Several of these capabilities now appear in both formats.

Delta Lake vs Apache Iceberg: how they compare

Capability	Delta Lake	Apache Iceberg
Governance	Independent project, a series of LF Projects, LLC under the Linux Foundation	Apache Software Foundation project
License	Apache License 2.0	Apache License 2.0
Latest release (October 2026)	4.4.1	1.12.0
Table state	Parquet data files plus a transaction log of versioned commits	Metadata files that track snapshots, manifest lists and manifests
Commits	Optimistic concurrency control; a conflict raises an exception and leaves the table intact	Writers swap the metadata file pointer atomically and retry if the base snapshot changed
Row-level deletes	Deletion vectors mark rows removed without rewriting Parquet files	Row-level delete files (format v2) and binary deletion vectors (v3)
Data layout	Liquid clustering, whose clustering columns can change without rewriting existing data	Hidden partitioning, with partition evolution as a metadata operation
Reading changes	Change data feed, which tracks row-level changes between table versions	The create_changelog_view procedure in Spark
Catalog	Filesystem-based by default; catalog-managed tables from Delta Lake 4.0.1	Tables created through a catalog or the Tables interface; the project defines a REST catalog API
Cross-format access	UniForm writes Iceberg metadata (and Hudi metadata, in preview) for Delta tables	The iceberg-delta-lake module snapshots Delta tables into Iceberg (documented for minReaderVersion 1 and minWriterVersion 2)

How to choose between them

• Who writes. Identify each engine that writes the table. Iceberg and Hudi clients get read-only access to UniForm tables, so a writing engine needs the format natively.
• Catalog. Check which formats your catalog manages and which clients it serves, such as Iceberg REST catalog clients or Delta catalog-managed tables.
• Features you rely on. Change data feed and liquid clustering are Delta Lake features. Hidden partitioning, partition evolution, and branches and tags are Iceberg features.
• Reader versions. Delta protocol upgrades are irreversible and can break older readers, and Iceberg raises its format version when older readers wouldn't read new features correctly. Confirm every engine before you enable new features.

How Databricks approaches Delta Lake and Apache Iceberg

Databricks documents support for both formats. Its docs recommend Unity Catalog managed tables for Delta Lake and Apache Iceberg alike, and Delta Lake is the format a new table gets unless you specify another. Databricks supports Iceberg tables that follow versions 1, 2 and 3 of the spec, and managed Iceberg became generally available in May 2026 (announcement). For Delta tables that Iceberg clients need to read, Iceberg reads (UniForm) add Iceberg metadata and leave the Parquet data files in place.

FAQs

Is Delta Lake still actively developed?

Yes. Delta Lake 4.4.1 was released on October 5, 2026, after 4.4.0 added Apache Spark 4.2 support in August. Databricks, which originally developed the Delta Lake protocol, continues to contribute to the open-source project.

Can the same table be read as Delta Lake and as Iceberg?

Yes. Delta Lake's UniForm generates Iceberg metadata asynchronously after a Delta commit, so Iceberg clients read the same Parquet files as Delta clients. That access is read-only for Iceberg clients.

Do I have to convert Delta tables to use Iceberg engines?

Not for reads, which UniForm covers. If an Iceberg engine must write the data, Iceberg documents a snapshotDeltaLakeTable action for supported Delta tables and full-copy options such as CREATE TABLE ... AS SELECT.

Which format does Databricks create by default?

Delta Lake. Databricks creates a Delta Lake table unless you specify another format, and Unity Catalog managed tables also support Apache Iceberg.