Companies running Apache Iceberg in production include Netflix, which built Iceberg in the first place to replace Hive tables that couldn’t hold up on S3, and has since moved its entire data warehouse, hundreds of petabytes, onto an Iceberg-only architecture, with teams across membership and finance, Cassandra data movement, Studio reporting, and media asset management all reading Iceberg’s own snapshot and partition metadata directly; Apple, whose Data Lakehouse team built GDPR- and DMA-compliant row-level deletes into Iceberg and Spark, running across tens of petabytes and tens of millions of files; Airbnb, which moved its Kafka-fed event-ingestion pipeline off Tez and Hive onto Spark 3 and Iceberg, consolidating hourly and daily tables via partition-spec evolution and cutting compute by more than 50%; Pinterest, whose Logging Platform and ML Infrastructure teams both rebuilt on Iceberg, cutting CDC ingestion latency from over 24 hours to minutes and ML feature backfill time from 140 days to 26; and LinkedIn, whose OpenHouse control plane governs more than 300,000 Iceberg tables holding over an exabyte of data, with automated compaction and cross-region replication built in. This page indexes production Apache Iceberg use cases, growing as we research and publish more.
Apache Iceberg use cases
Netflix
Netflix runs Apache Iceberg across its entire data warehouse, from an exabyte-scale migration off Hive tables to Maestro’s Incremental Processing Solution, the Psyberg framework for late-arriving membership and finance data, and its Cassandra-to-Iceberg data movement engine. Sourced from Netflix’s own engineering blog, an AWS re:Invent conference talk, and a public Netflix migration tool on GitHub.
Apple
Apple built GDPR- and DMA-compliant row-level deletes into Iceberg and Spark, running production tables holding tens of petabytes of data and tens of millions of files, alongside storage-partitioned joins and a shared multi-engine deployment spanning Spark, Trino, and Flink. Sourced from a peer-reviewed paper co-authored by eight Apple engineers, a keynote and interview with the engineering manager who led the work, and two further conference talks by named Apple engineers.
Airbnb
Airbnb moved its Kafka-fed event-ingestion pipeline off Tez and Hive onto Apache Spark 3 and Iceberg, consolidating separate hourly and daily tables into one via partition-spec evolution and cutting compute by more than 50% against the prior stack. Sourced from Airbnb’s own engineering blog, a design discussion posted to the Apache Iceberg developer mailing list, and a Trino Summit talk by two named Airbnb engineers.
Pinterest rebuilt both its database CDC ingestion pipeline and its ML feature backfill infrastructure on Apache Iceberg, cutting CDC pipeline latency from over 24 hours to minutes and ML feature backfill time from 140 days to 26, alongside a Ray-based feature-joining layer built directly on Iceberg-bucketed tables. Sourced from four Pinterest Engineering Blog posts and Pinterest’s own internal Iceberg fork on GitHub.
LinkedIn built OpenHouse, a declarative control plane for Apache Iceberg, that now governs more than 300,000 tables holding over an exabyte of data, with automated compaction, cross-region replication, and governance applied by default. Sourced from LinkedIn’s own engineering blog, the official linkedin/openhouse GitHub repository, a peer-reviewed SIGMOD paper co-authored with Microsoft and University of Maryland researchers, and conference talks by the engineer who has led the work since 2023.
FAQ
What companies use Apache Iceberg in production?
Netflix, which created Apache Iceberg, runs it across its entire data warehouse, as detailed in our own use-case writeup on Netflix. Apple built GDPR- and DMA-compliant row-level deletes into Iceberg and Spark, running production tables at tens of petabytes and tens of millions of files, as detailed in our own use-case writeup on Apple. Airbnb moved its Kafka-fed event-ingestion pipeline off Tez and Hive onto Spark 3 and Iceberg, cutting compute by more than 50%, as detailed in our own use-case writeup on Airbnb. Pinterest rebuilt both its database CDC ingestion pipeline and its ML feature backfill infrastructure on Iceberg, as detailed in our own use-case writeup on Pinterest. LinkedIn built OpenHouse, a declarative control plane now governing over 300,000 Iceberg tables holding more than an exabyte of data, as detailed in our own use-case writeup on LinkedIn. We are adding further named production deployments to this page as we research and publish them.
What is Apache Iceberg used for?
Apache Iceberg is used as a table format for large analytic datasets stored in an object store like S3, adding schema evolution, hidden partitioning, snapshot isolation, and time travel to data that would otherwise need a proprietary or Hive-based warehouse.