Power BI Direct Lake: fast reports on large OneLake data
Direct Lake in Power BI is a table storage mode for semantic models in Microsoft Fabric. It loads data into memory straight from Delta tables in OneLake, so a model can serve large volumes without a full copy on every refresh. This guide explains how the mode works, how it compares with Import and DirectQuery and which requirements Microsoft documents. In short, Direct Lake reads Delta tables in OneLake without a scheduled import, and some models can fall back to DirectQuery when a query cannot run in Direct Lake.
- Reads Delta tables
- Refresh copies metadata
- Needs Fabric capacity
Last updated
What is Power BI Direct Lake?
Microsoft describes Direct Lake as a storage mode option that is optimized for large volumes of data. The data loads quickly into memory from Delta tables in OneLake, the single store for analytics data in Fabric. After that, the model supports fast interactive analysis.
The mode shares its engine with Import. Both process queries with the VertiPaq engine, whereas DirectQuery sends each query to the underlying source. Therefore, according to Microsoft, Direct Lake and Import normally outperform DirectQuery when readers load and use report visuals.
Only the needed data enters memory. Microsoft notes that the analyzed volume can exceed the memory limit of the capacity for this reason. So the mode suits lakehouses and warehouses where a full Import copy is impractical.
How does a Direct Lake refresh differ from Import?
An Import refresh replicates the data. It builds a complete cached copy for the model, which can take considerable time and use significant memory and processing power on the source and the capacity.
A Direct Lake refresh copies only metadata instead. Microsoft calls this step framing. It analyzes the latest version of the Delta tables and updates the references to the newest files in OneLake. As a result, the operation is low in cost and can finish in a few seconds.
Data preparation moves upstream as well. With Direct Lake, the work happens in OneLake with tools such as Spark jobs, T-SQL statements, dataflows and pipelines. Import mode, however, still supports preparation with Power Query inside the semantic model.
Direct Lake, Import and DirectQuery compared
This summary draws on the comparison table in Microsoft's overview article.
| Point | Direct Lake | Import | DirectQuery |
|---|---|---|---|
| License | Fabric capacity only | Any Fabric or Power BI license | Any Fabric or Power BI license |
| Query engine | VertiPaq | VertiPaq | The source database |
| Refresh | Copies metadata | Copies all data | No data copy |
| Data source | Fabric items with Delta tables | Any connector | Connectors that support it |
| Hybrid tables | No | Yes | Yes |
| User-defined aggregations | No | Yes | Yes |
Three terms to know before you start
Microsoft uses these terms throughout its documentation, so it helps to learn them first.
Direct Lake on OneLake
This option can use Delta tables from one or more Fabric sources. It does not fall back to DirectQuery, and you can add Import tables.
Direct Lake on SQL
This option uses a single Fabric source through its SQL analytics endpoint. It can fall back to DirectQuery, for example for a SQL view.
Guardrails
Each capacity size sets limits per table, such as the number of rows and files. Therefore check the table for your size.
When is Direct Lake the right choice?
Microsoft names the primary use case plainly. Direct Lake fits analytics projects led by IT that follow a lake-centric architecture. In such projects, large volumes of data already sit in OneLake or will gather there.
Import and DirectQuery remain relevant, however. Microsoft states that Import often works well for a self-service analyst who needs freedom to act quickly without depending on IT. A model can also combine Import tables with Direct Lake tables.
Table quality matters too. Microsoft points out that performance depends on well-tuned Delta tables. It also recommends a prototype or proof of concept first, to confirm that the mode is the right solution and to reduce risk. If you are still choosing between storage modes, see Power BI DirectQuery vs Direct Lake. To connect the lakehouse that holds the tables, see Power BI Fabric lakehouse integration, and for wider platform planning, see Microsoft Fabric consulting.
Six documented limits of Power BI Direct Lake
Microsoft says these features evolve rapidly, so check the current article before you decide.
- Models require a Fabric capacity license
- Personal workspaces cannot hold Direct Lake models
- Refresh through a data gateway is not supported
- Complex Delta column types are not supported
- Related columns in a relationship need matching data types
- The model and its source must be in the same region
Power BI Direct Lake: common questions
Does Power BI Direct Lake need a special license?
Yes. Microsoft states that Direct Lake semantic models require a Fabric capacity license. Import and DirectQuery, in contrast, work with any Fabric or Power BI license. Capacity guardrails also apply, and they differ by capacity size. So confirm the limits for your size before you plan a model.
What happens when a table exceeds the guardrails?
The behavior depends on the option. With Direct Lake on OneLake, the refresh fails, and nobody can query the model until you optimize the Delta tables. With Direct Lake on SQL, queries fall back to DirectQuery if fallback is enabled. They still return results, but more slowly.
Can a Direct Lake model use row-level security?
Yes. Microsoft lists row-level security in the semantic model as available for both options, and it strongly recommends a fixed identity cloud connection with it. Security rules defined at the SQL analytics endpoint behave differently, however. Direct Lake on OneLake does not apply them, so plan security in the model.
Can I edit a Direct Lake model in Power BI Desktop?
Yes. According to Microsoft, the desktop app can live edit a semantic model with Direct Lake tables and Import tables. Web modeling in the service can open such models as well. Some checks are limited there, because the editor assumes that your relationship settings are correct.
Can Power BI Agency help us assess Direct Lake?
Yes. One of our consultants reviews your Fabric sources, your capacity size and your reporting needs, then builds a small proof of concept with you. We work online inside your own Microsoft tenant and do not resell licenses. Power BI Agency is a division of Sataware Technologies and is independent of Microsoft.
Let's build your first dashboard together.
Tell us what you report on today. A consultant replies within 2 hours with next steps.
Trusted partners
Microsoft Solutions Partner · AWS Partner
Cloud & AI Platforms
Azure SQL, Microsoft Fabric and the data pipelines behind your reports.
AI Business Solutions
Power BI reporting on your Dynamics 365, Business Central and Power Platform data.
Security
Row-level security and Entra ID sign-in on every report and workspace.
Amazon Web Services
Power BI on your AWS data, with connectors for Redshift, Athena and S3.