Iceberg Summit 2026 Draws All Major Cloud Vendors as V3 Adoption Accelerates Across the Data Lakehouse Ecosystem
The third Iceberg Summit in San Francisco brought 70+ sessions and announcements from Snowflake, Google, Dremio, Databricks, and AWS as the open table format's V3 specification reaches production readiness.
Overview
The Apache Iceberg community gathered in San Francisco on April 8-9 for its third annual Iceberg Summit, drawing more than 70 sessions and participation from every major cloud data platform vendor. The two-day event at the Marriott Marquis became the stage for a wave of announcements centered on the Iceberg V3 specification, with Snowflake, Google Cloud, Dremio, Databricks, and AWS all unveiling expanded support for the open table format that has become the de facto standard for data lakehouse architectures.
The summit underscored a notable shift: the competitive battleground in enterprise data infrastructure has moved from proprietary formats to who can offer the best interoperability and governance on top of a shared, open standard.
What We Know
Iceberg V3 Reaches Critical Mass
The V3 specification, which the Apache Iceberg project finalized in 2025, introduces deletion vectors for efficient row-level deletes without rewriting entire data files, the VARIANT data type for semi-structured data, row-level lineage tracking for change data capture and auditing, geospatial data types, and nanosecond-precision timestamps. All five major cloud platforms now offer some level of V3 support, marking the fastest adoption cycle for an Iceberg spec version to date.
Snowflake placed its V3 support in public preview in early March, covering deletion vectors, row lineage, and the VARIANT type across its streaming and batch ingestion pipelines. At the summit, the company announced further plans to bring V3 to general availability, according to SiliconANGLE. James Rowland-Jones, Snowflake’s director of product management, stated: “What’s new here is the expansion from foundational interoperability to more complete, production-ready interoperability.”
Databricks, which had previously announced Iceberg V3 features in Databricks Runtime 18.0, made Unity Catalog-managed Iceberg tables available in public preview as of April 7, one day before the summit opened.
Google Announces BigQuery-Iceberg Read/Write Interoperability
Google Cloud used the summit to announce the preview of full read and write interoperability between BigQuery and Iceberg-compatible engines, according to the company’s blog post. The integration runs through Google’s managed Iceberg REST Catalog, allowing teams to create, update, and query Iceberg tables using BigQuery alongside Spark, Flink, Trino, and other engines.
The announcement includes table-level access controls via credential vending, enabling uniform governance across multiple compute engines that query or modify the same Iceberg tables. Google said the features will be generally available for BigQuery-managed Iceberg tables in the coming weeks.
Dremio Highlights V3 Support and Polaris Momentum
Dremio, one of the earliest commercial backers of Apache Iceberg, announced that its cloud platform now provides full read and write support for Iceberg V3, including deletion vectors for accelerated CDC workloads, the VARIANT type, and row-level lineage, as reported by Yahoo Finance.
The company also highlighted the graduation of Apache Polaris, the open-source Iceberg catalog that Dremio co-founded, to a top-level Apache Software Foundation project. Dremio engineer JB Onofre, who guided Polaris through incubation, was elected to the ASF board. Polaris now supports full read and write access from any REST-compatible engine, including Spark, Flink, Trino, and DuckDB.
Snowflake Pushes Governance Portability
Beyond V3, Snowflake used the summit to outline a broader governance portability strategy, according to SiliconANGLE. The company is promoting Apache Polaris as the mechanism for making governance policies portable across systems, implementing policy exchange standards, governance federation, and read restriction APIs that allow one system to share pre-evaluated access rules without copying data.
Snowflake also introduced pg_lake, a PostgreSQL extension that enables direct querying of data lake formats such as Parquet and CSV, and can write to Iceberg tables without ETL. The company additionally announced investment in OpenLineage and the Open Semantic Interchange specification, backed by more than 35 industry partners, to standardize business definitions for AI systems.
Keynotes Frame the Format’s Future
The summit’s two keynotes reflected the format’s expanding ambitions. Russell Spitzer, an Apache Iceberg PMC member and principal software engineer at Snowflake, delivered a keynote titled “From Batch to Streaming and AI: Iceberg for Everyone by Everyone,” signaling the community’s push to extend the format beyond traditional batch analytics. Julien Le Dem, Apache Parquet PMC Chair and principal software engineer at Datadog, spoke on “Column Storage for the AI Era,” addressing how columnar formats must evolve to serve machine learning and AI workloads.
What We Don’t Know
The V3 specification is widely available in preview, but production readiness varies. Snowflake’s public preview, for instance, does not yet support row-level equality deletes or certain table properties, and externally-managed tables remain limited to V2 writes. Similar limitations exist across other platforms. How quickly enterprises will migrate existing V2 tables to V3 in production remains an open question, particularly given that V3 upgrades are one-way and cannot be downgraded.
The governance portability proposals from Snowflake and the broader Polaris ecosystem are still early. Whether competing vendors will adopt shared governance standards or maintain proprietary approaches could determine whether Iceberg delivers on its promise of true multi-engine, multi-cloud data interoperability.
The summit’s emphasis on AI workloads also raises questions about whether the Iceberg format, designed primarily for analytical queries, can efficiently serve the emerging patterns of retrieval-augmented generation, vector search, and continuous model training that increasingly define enterprise data infrastructure.
Analysis
The convergence of all major cloud vendors on Iceberg V3 at a single event marks a significant milestone for the open data lakehouse movement. Three years ago, the landscape was fractured between Delta Lake, Iceberg, and Apache Hudi. The 2026 summit suggests that Iceberg has won the format war, at least at the specification level, with even Databricks, the creator of Delta Lake, offering first-class Iceberg support through Unity Catalog.
The competitive differentiation is shifting from format support to the services built around it: catalog management, governance, query optimization, and AI integration. Snowflake’s pg_lake and governance portability push, Google’s BigQuery interoperability, and Dremio’s autonomous reflections all represent different bets on where the value will accrue in an open-format world.
For data engineers, the practical takeaway is that V3 adoption is no longer a question of whether but when. The deletion vectors, VARIANT type, and row lineage features address long-standing pain points in data lakehouse operations, and the broad platform support means migration paths are available regardless of cloud provider.