Direct Lake
A storage mode where the semantic model reads Delta tables in OneLake directly, with no import refresh.
Direct Lake promises Import-mode speed without the import refresh. It delivers, as long as the tables underneath are built for it. When they are not, it quietly gets slower.
Part of the Model layer of the ecosystem. Reviewed 13 November 2026 by Piyush Jain.
What Direct Lake is
In Import mode, Power BI copies data into the model on every refresh. In Direct Lake, the model loads columns straight from the Delta tables in OneLake when a query needs them. A refresh only updates which version of the tables the model points to, which takes seconds.
There are two flavours. Direct Lake on SQL endpoints can fall back to DirectQuery when it cannot read a table directly, for example when the source is a view or the table is over the capacity's limits. Direct Lake on OneLake does not fall back. Each Fabric SKU sets guardrails, such as rows per table and model size in memory.
When to use it
Use it when
- Your data already lands in a Lakehouse or Warehouse as Delta tables.
- Import refreshes are long, or the data must be fresh several times a day.
- The tables fit within the guardrails of your Fabric SKU.
Think twice when
- Your sources are not in Fabric yet. Import mode is simpler until they are.
- The model needs calculated columns or tables on the Direct Lake tables, or Power Query steps. Those belong upstream in the Lakehouse.
The mistakes I see most
From opening other people's models. Each one is fixable.
- 1
Assuming it is always faster
The first query after a reframe has to load columns into memory, so it is slower than Import. Warm-up matters, and so does keeping only the columns reports use.
- 2
Reading views and causing fallback
A model on a SQL endpoint view falls back to DirectQuery on every query. Reports work, so nobody notices, until they ask why Fabric is slower than the old Premium setup.
- 3
Ignoring the guardrails
A fact table over the row limit for your SKU also triggers fallback, or fails to load. Check the guardrails before you choose the SKU, not after go-live.
Questions people ask
- Is Direct Lake faster than Import mode?
- Once columns are in memory, query speed is similar to Import. The gains are in refresh: no long import, and fresher data. The first queries after a refresh can be slower while columns load.
- What makes Direct Lake fall back to DirectQuery?
- With Direct Lake on SQL endpoints: reading a view instead of a table, tables over the SKU's guardrails, or security defined in the SQL endpoint. Fallback keeps reports working but slower.
- Do I need Fabric capacity for Direct Lake?
- Yes. Direct Lake needs the model and its data on Fabric capacity (an F SKU) or a Power BI Premium P SKU.
Sources: Microsoft Learn: Direct Lake overview. Opinions are my own.