Momentum One Academy · glossary
Fabric Data Warehouse
Definition
Fabric Data Warehouse is a SQL-first relational warehouse item in Microsoft Fabric. It stores data on OneLake in Delta tables and exposes a relational surface primarily developed with T-SQL. It suits structured and semi-structured data shaped into star or snowflake schemas, curated marts, and governed semantic models. The Warehouse supports data-definition and data-manipulation operations, including multi-table ACID transactions. 1
Where it sits in Fabric
A Warehouse is a workspace item alongside Lakehouses, Data Factory pipelines, and semantic models. OneLake is the shared foundation; both Warehouse and Lakehouse use Delta, but a Lakehouse favors file-oriented engineering and Spark, while its autogenerated SQL analytics endpoint is read-only for data changes. A Warehouse is the natural home when SQL is the primary development and serving experience. 1 2
A semantic model can sit above Warehouse tables. Direct Lake can consume supported Warehouse Delta tables through OneLake, so it is a semantic-model storage mode, not another warehouse or a replacement for upstream transformations. 3
When to use it
Choose a Warehouse when a solution needs a SQL-led dimensional model, governed business definitions, set-based transformations, and transactional loading across related tables. Choose a Lakehouse when highly unstructured data and Spark-led engineering dominate. Use SCD Type 2 selectively inside dimensions when historical perspective matters: expire the current version and insert a new version with a surrogate key and validity boundaries rather than overwriting the past. 1 4 5
Practical example
For retail, a pipeline lands orders and customer extracts in staging tables. T-SQL loads FactSales, DimDate, DimProduct, and DimCustomer. If customer C-123 moves from the West region to the East region, the load closes the current dimension row, inserts a new row with a new surrogate key, and sets ValidFrom, ValidTo, and IsCurrent. Facts resolve the version valid for the sale, preserving historical reporting. A Direct Lake semantic model can expose the star schema to Power BI without importing a second full copy of the data. 3 4
Production caution
Do not truncate and fully reload an SCD Type 2 dimension that uses generated surrogate keys. That can invalidate fact-table references and make historical versions impossible to reconstruct safely. Use an incremental, auditable expire-and-insert process, and test late-arriving changes and soft-delete rules before publishing reports. 5
Module 7 connection
In Module 7, treat the Warehouse as the governed serving layer that turns Lakehouse or other landed data into a relational model. Connect SCD Type 2 to dimension loading, then connect the finished Warehouse tables to a Direct Lake semantic model for consumption. Fabric can use Lakehouse for engineering, Warehouse for SQL-first curation, and Direct Lake for semantic-model access over the resulting Delta data.
Ready to apply this design in a production-minded Fabric solution? Explore Momentum One Academy’s Fabric Production Workshop.
References
Plan your workshop