Databricks 2,717 set out the lakehouse idea in its CIDR 2021 paper (Where Warehouses Strained) and still defines its commercial form. Tables are Delta Lake 237,929 by default and, on Databricks Runtime 16.4 LTS or later, can also be managed Iceberg 129 tables; both live in your own cloud storage and are governed by Unity Catalog, the commercial sibling of Unity Catalog OSS's Unity Catalog OSS. Unity Catalog grants privileges down to rows and columns, records lineage, and exposes every Iceberg table through the Iceberg REST Catalog API with credential vending, so Trino 403,499 , DuckDB 61,228 or Snowflake can read and write them (the pattern of Lakehouse Access Control, as a service). Delta tables reach Iceberg readers through UniForm (UniForm), and Iceberg tables managed elsewhere (Glue, Hive 129 , Snowflake) appear read-only through Lakehouse Federation.
Maintenance is predictive optimization: Databricks decides when to run OPTIMIZE (compaction and clustering), VACUUM (deleting unreferenced files) and ANALYZE (statistics) on managed tables, the jobs Iceberg in Production scheduled by hand. It is on by default for accounts created since 11 November 2024 and runs on serverless jobs compute, billed in DBUs like everything else (Databricks and DBUs). Storage is billed by your cloud provider.
The trade-off is the usual one. Unity Catalog's open API and open table formats keep the data portable, but the features that make the platform pleasant (Photon, predictive optimization, serverless warehouses) exist only there, and its DBU bill is the hardest of the five to forecast without a trial.