AI glossary · Advanced · Data engineering · AI platforms & solutions
Data lakehouse
A data lakehouse is a data storage and analytics architecture that combines in one system the strengths of a 'data lake,' which cheaply stores anything in its original form, and a 'data warehouse,' which analyzes organized data quickly and reliably.
About this term
- Level
- Advanced · Specialist terms that need the inner workings
- Category
- Data, ops & platforms
- Added
- 2026-10-08
What's different
A data lake stores all data, such as logs, images, documents and tables, in its original format in cheap cloud storage. It is flexible, but if management gets loose it easily becomes a 'data swamp' where no one knows what is where. A data warehouse is well organized but has trouble handling unstructured data and is costly. So in the past, companies kept both and copied data between them; a lakehouse merges the two into one place.
Lake·warehouse·lakehouse
| Unstructured data | Organization·reliability | Low-cost storage | |
|---|---|---|---|
| Data lake | ✓ | ✕ | ✓ |
| Data warehouse | ✕ | ✓ | △ |
| Lakehouse | ✓ | ✓ | ✓ |
How it works
At the bottom, data is stored in cloud object storage in open file formats such as Parquet. On top of that sits a 'table format' layer such as Delta Lake, Apache Iceberg or Apache Hudi, which lets a collection of files be handled like a database table. This layer provides warehouse-like features such as change history, protection against conflicts when several writers write at once, and querying data as it was at a past point in time. SQL analytics tools and machine learning training tools can then read the same storage together without copying the data.
Layers of a lakehouse
Analytics and AI toolsSQL analytics, BI, machine learning training
Table format layerDelta Lake·Iceberg·Hudi, change history·conflict prevention
Cloud object storageRaw data stored in open file formats such as Parquet
Why it's getting attention
In the AI era, unstructured data such as documents, images and conversation logs, not just tabular data, has become a core ingredient for training and RAG. A lakehouse manages this kind of data in one place alongside existing sales data, so analytics teams and AI teams can work from the same data. Because data is not copied many times, cost and inconsistency problems also go down.
Things to know
Because a table format, storage and processing engines have to be combined, initial design and operation require expertise. For some reports that need very fast responses, a dedicated warehouse may still be more convenient. Using open formats reduces lock-in to a specific vendor, but access control and data governance still need to be handled carefully on their own.
Example
A financial company stores transaction records (tables), transcribed text of customer service calls and images of documents submitted by customers all in a cloud lakehouse. The analytics team pulls monthly transaction statistics with SQL, the AI team trains a customer inquiry classification model on the call text in the same storage, and the internal chatbot uses content extracted from the documents for RAG search. Each department does not need to keep its own copy of the data.