Are your Power BI reports taking too long to load? Do you use DirectQuery and find your visuals freezing with every filter? There's a connection mode that promises to change that — and it's already available for Microsoft Fabric users.
Direct Lake Mode is the new performance standard in Microsoft's Analytics ecosystem. In this article, you'll understand how it works, when it makes sense to use it, and how it compares to the modes you already know.
The Three Power BI Connection Modes
To understand Direct Lake, we first need to revisit what already exists.
Import Mode
The most popular mode. Data is copied into Power BI, compressed in memory by the VertiPaq engine, and queried at very high speed. The catch? Data is only updated when you schedule a refresh — and very large datasets can take hours to load.
Use when: the dataset is small to medium-sized, data doesn't need real-time updates, and interaction speed is the top priority.
DirectQuery
In this mode, Power BI stores nothing. Every click on a visual generates a direct query against the data source. Data is always up to date, but performance depends entirely on how fast the database responds — and that can be slow.
Use when: data must always be in real time, or the volume is too large to import.
Direct Lake Mode
Here's the star of the show. Direct Lake was designed to be the best of both worlds: the speed of Import Mode combined with the near real-time freshness of DirectQuery — without duplicating data.
Use when: you operate within Microsoft Fabric with Delta tables on OneLake and need high performance with always-current data.
How Direct Lake Works in Practice
Direct Lake neither imports data nor queries the original source on every click. Instead, it connects the Power BI semantic model directly to Delta tables stored on OneLake — Microsoft Fabric's unified storage layer.
The process works like this:
- Data lives on OneLake, stored in Parquet format, which is efficient for columnar queries.
- When a query runs, the VertiPaq engine loads only the columns needed into memory — not the entire dataset.
- When the data changes at the source, the Direct Lake refresh updates only the metadata. This takes seconds, not minutes or hours.
The result? Report response times close to Import Mode, with data that can reflect changes in near real time.
Direct Lake on OneLake vs. Direct Lake on SQL
Since March 2026, Direct Lake on OneLake has been generally available and is Microsoft's recommended option.
- Better compatibility with OneLake security
- More modeling features, including calculated columns introduced in April 2026
- Superior query performance
Direct Lake on SQL uses the Fabric SQL analytics endpoint and allows fallback to DirectQuery — useful in specific permission and governance scenarios.
When Direct Lake Falls Back to DirectQuery (and What to Do About It)
An important note: Direct Lake can automatically fall back to DirectQuery mode in some situations.
- Highly complex DAX queries
- High-cardinality relationships
- Fabric capacity memory limits
- Using views instead of physical tables
This is called fallback, and it can cause a noticeable drop in performance. To avoid it:
- Use physical tables rather than views in the Lakehouse
- Optimize Delta tables with V-Order, already enabled by default in Fabric
- Right-size Fabric capacity for expected data volume and concurrency
Quick Comparison: Import vs. DirectQuery vs. Direct Lake
- Data stored — Import Mode: in Power BI memory | DirectQuery: at the original source | Direct Lake: in OneLake (Delta)
- Refresh — Import Mode: scheduled | DirectQuery: real time | Direct Lake: near real time
- Performance — Import Mode: high | DirectQuery: variable | Direct Lake: high
- Requires Fabric — Import Mode: no | DirectQuery: no | Direct Lake: yes
- Best for — Import Mode: medium datasets | DirectQuery: real-time data | Direct Lake: large volumes in Fabric
Is It Worth Migrating to Direct Lake?
If your organization already uses Microsoft Fabric — or is planning to migrate — the answer is almost always yes. Direct Lake reduces refresh costs, eliminates data duplication, and delivers Import Mode-equivalent performance even at large scale.
But if you still operate outside of Fabric, Import Mode remains the best choice for most scenarios. And DirectQuery is still necessary when data must be absolutely real time and the data source handles query load well.
The good news is that Direct Lake isn't a radical replacement — it's a natural evolution for teams growing within the Microsoft ecosystem.
Conclusion
Direct Lake Mode represents a real shift in how Power BI handles data at scale. By combining the speed of VertiPaq with the flexibility of OneLake, Microsoft resolved one of the platform's biggest trade-offs: choosing between performance and freshness.
If you work with Analytics in a Fabric environment, this mode is worth studying carefully — and it's time to start planning a gradual migration of your most critical solutions.