A data architecture paradigm that combines the best features of data lakes and data warehouses, providing ACID transactions, schema enforcement, and governance on low-cost storage with support for both business intelligence and machine learning workloads. Effective implementation supports data-driven strategies and helps maintain data integrity across systems.
Lakehouse Architecture
Lakehouse Architecture is a data management pattern that merges the low-cost, open storage of a data lake with the transactional guarantees and governance of a data warehouse. It puts ACID transactions, schema enforcement, and fine-grained access control directly on top of inexpensive object storage, so a single copy of the data can serve both business intelligence and machine learning workloads. The approach, popularized by Databricks around 2020, is built on open table formats like Delta Lake, Apache Iceberg, and Apache Hudi.
- Open table formats - Delta Lake, Iceberg, and Hudi add transaction logs and metadata over columnar files like Parquet.
- ACID transactions - Reliable concurrent reads and writes that a raw data lake cannot guarantee on its own.
- Schema enforcement and evolution - Warehouse-grade data quality without giving up the lake’s flexibility.
- One copy, many engines - BI, SQL analytics, and ML/AI training all read the same governed tables.
- Decoupled storage and compute - Cheap object storage scales independently of the query engines that touch it.
For API operations, the lakehouse is increasingly the system of record that data and analytics APIs sit in front of. Catalog and query services expose lakehouse tables through REST, SQL, and increasingly agent-facing endpoints, while governance and lineage metadata become the contract that tells consumers what a dataset means and who may use it. As AI agents reach for governed enterprise data, the lakehouse’s open formats and central access controls make it a natural backend for retrieval and grounding without copying data into yet another silo.
