Data lakehouse
Also: Lakehouse, Data lakehouse, Maison du lac de donnees, Lac-entrepot de donnees
A hybrid architecture combining the flexibility of a data lake with the analytical capabilities of a data warehouse, on a single storage layer.
What it is
A data lakehouse is a data architecture that merges two worlds that used to be separate:
- The data lake: cheap, scalable storage (typically object storage like Amazon S3, Azure Data Lake Storage, or Google Cloud Storage) that holds raw data in any format (structured tables, JSON, images, logs, video).
- The data warehouse: a governed, high performance system optimized for SQL analytics, with schemas, transactions, and fast queries.
The lakehouse keeps all data in open file formats (such as Parquet) on low cost storage, then adds a metadata and transaction layer on top (open table formats like Apache Iceberg, Delta Lake, or Apache Hudi). This layer brings warehouse-grade features (ACID transactions, schema enforcement, versioning, and governance) directly to the lake.
Why it matters
Before the lakehouse, organizations often ran two stacks: a lake for data science and raw storage, and a separate warehouse for business intelligence. Data was copied between them, which meant duplicated storage, extra pipelines, higher cost, and multiple versions of the truth.
The lakehouse promises:
- One copy of the data serving BI dashboards, ad hoc SQL, and machine learning.
- Lower cost by using object storage instead of proprietary warehouse storage.
- Open formats that avoid vendor lock-in.
- Governance and quality applied once, consistently.
How it is used in practice
Teams typically organize the lakehouse in tiers, often called bronze, silver, gold:
- Bronze: raw ingested data, as received.
- Silver: cleaned, deduplicated, conformed data.
- Gold: business-ready aggregates and metrics for reporting.
Analysts query gold tables with SQL. Data scientists train models on silver or bronze data. Both use the same underlying files.
Concrete worked example
A retailer streams point of sale transactions, website clickstream, and inventory feeds into object storage (bronze). A pipeline cleans and joins them into silver tables using an open table format that guarantees a consistent snapshot even while new data lands.
- The finance team queries gold tables for daily revenue and margin.
- The marketing team builds customer segments from the same clickstream.
- The data science team trains a churn model on the raw event history.
No data is copied into a separate warehouse. One governed platform, many consumers.
See also
Frequently asked questions
What is a data lakehouse in simple terms?
A data lakehouse is an architecture that stores all data in open file formats on cheap object storage (Amazon S3, Azure Data Lake Storage, Google Cloud Storage) and adds a metadata and transaction layer on top so it behaves like a data warehouse. That layer brings ACID transactions, schema enforcement, versioning and governance directly to the lake. The result: one copy of the data serves BI, ad hoc SQL and machine learning.
What is the difference between a data lake, a data warehouse and a lakehouse?
A data lake is cheap, scalable storage for raw data in any format (tables, JSON, images, logs, video) but without warehouse guarantees. A data warehouse is a governed, high performance system optimized for SQL analytics, with schemas, transactions and fast queries. The lakehouse keeps the lake's storage and open formats, then layers warehouse features on top, so you no longer need to maintain two separate stacks.
Why would a company move away from running a lake and a warehouse side by side?
Because two stacks mean copying data between them: duplicated storage, extra pipelines, higher cost and several versions of the truth. The lakehouse keeps one governed copy of the data, uses object storage instead of proprietary warehouse storage, relies on open formats that limit vendor lock-in, and applies governance and quality rules once rather than twice.
What do bronze, silver and gold mean in a lakehouse?
They are the three tiers teams use to organize data in a lakehouse. Bronze holds raw ingested data as received, silver holds cleaned, deduplicated and conformed data, gold holds business-ready aggregates and metrics for reporting. Analysts query gold tables in SQL while data scientists train models on silver or bronze, all on the same underlying files.
Which technologies make a lakehouse possible?
Two building blocks: an open file format such as Parquet for the data itself, and an open table format such as Apache Iceberg, Delta Lake or Apache Hudi for the metadata and transaction layer. The table format is what guarantees ACID transactions, schema enforcement and consistent snapshots, so a pipeline can read a stable view of a table even while new data is landing.