Momentum One Academy · glossary
Direct Lake
Definition
Direct Lake is a Power BI semantic-model table storage mode in Microsoft Fabric. It reads Delta tables stored in OneLake and loads only the columns needed by a query into the VertiPaq in-memory engine. Unlike Import mode, a Direct Lake refresh normally updates metadata references to the latest Delta files (called framing) instead of building a complete second copy of the data. 1 2
Where it sits in Fabric
Direct Lake sits between Fabric storage and the semantic model: a lakehouse, warehouse, or other supported Fabric item prepares and stores Delta data in OneLake; the semantic model exposes business relationships, measures, and security to Power BI reports. OneLake remains the shared data foundation, while Direct Lake is the access mode used by the Analysis Services engine. Data preparation therefore belongs upstream in notebooks, pipelines, dataflows, or SQL rather than inside a large model refresh. 1 3
When to use it
Use Direct Lake when a Fabric solution has large, curated Delta tables, needs responsive interactive analysis, and cannot reasonably duplicate the full dataset in an imported model. It is a strong fit for a gold reporting layer and for data that changes often. Import may still be better for small self-service models or cases where the author needs Power Query transformations inside the semantic model. 1
Practical example
A lakehouse team transforms daily sales events into a partitioned Sales Delta table and publishes a star-schema semantic model with date, product, and customer dimensions. A report user filters sales by region. Direct Lake loads the required columns into memory, while a framing operation makes newly committed Delta files visible without a full data copy.
Production caution
Do not treat Direct Lake as automatically real time. Queries reflect the latest successful framing point, and large or poorly maintained Delta tables can hit capacity guardrails. Optimize table layout, monitor framing and automatic updates, validate OneLake or semantic-model permissions, and test whether a SQL-endpoint model falls back to DirectQuery, which can be slower. Direct Lake on OneLake does not support that fallback. 1 2 4
Workshop connection
In Module 8, connect the lakehouse and OneLake foundations to a production semantic model. Compare Direct Lake with Import and DirectQuery, then validate the model’s refresh behavior, table optimization, capacity, and security before releasing reports.
Ready to turn the pattern into a governed delivery? Explore the Momentum One Academy Fabric Production Workshop.
References
Plan your workshop