DirectLake Mode: Power BI Without the Refresh Schedules

The refresh problem

Power BI has historically required a trade-off between performance and data freshness. Import mode delivers fast query performance but requires scheduled refreshes — meaning data is always somewhat stale. DirectQuery gives you real-time data but at the cost of query performance, as every interaction hits the source system.

For organisations that need both freshness and performance — and most do — this trade-off has been a persistent frustration. Reports are either fast but stale, or fresh but slow.

What is DirectLake mode?

DirectLake mode is a new Power BI storage mode introduced with Microsoft Fabric. It reads Delta tables directly from OneLake — Fabric's unified storage layer — without importing data into the Power BI model and without routing queries through a SQL endpoint.

This is a fundamental architectural shift. DirectLake combines the query performance of import mode with the data freshness of DirectQuery. Data is always current because it's read directly from the Delta table, and queries are fast because they leverage Power BI's Vertipaq engine with direct access to the parquet files.

How it works

In DirectLake mode, Power BI creates a semantic model that maps directly to Delta tables in OneLake. When a user opens a report, the Vertipaq engine reads the parquet files directly — no data copy, no refresh schedule, no SQL translation layer.

Key technical points:

• No refresh needed — data is always current because it's read directly from the Delta table at query time
• Vertipaq performance — the same columnar engine that powers import mode, delivering sub-second query response on well-designed models
• Fallback to DirectQuery — if a query can't be served by DirectLake (e.g. complex DAX or unsupported functions), Power BI automatically falls back to DirectQuery against the SQL endpoint
• Partition awareness — DirectLake reads only the partitions needed for a query, making it efficient even on large tables

When to use DirectLake

DirectLake is ideal when:

• Your data is stored as Delta tables in OneLake or Fabric
• You need near-real-time data freshness in your Power BI reports
• Your semantic models are designed with best practices — star schemas, appropriate relationships and efficient DAX
• Your data volumes are large enough that import mode refreshes are slow or disruptive

When to be cautious

DirectLake may not be the right choice when:

• Your data isn't in Delta format — DirectLake requires Delta tables in OneLake, so you'll need to migrate your storage layer first
• Your reports rely on complex DAX that can't be served by DirectLake — these queries will fall back to DirectQuery, potentially degrading performance
• You need write-back or calculated tables that require import mode semantics
• Your Fabric capacity is undersized — DirectLake queries consume capacity units, so ensure your capacity can handle the workload

Getting started

To use DirectLake mode, your data must be in Delta format in a Fabric lakehouse or warehouse. Create a semantic model in Fabric and select DirectLake as the storage mode. Fabric will automatically map your Delta tables to the semantic model.

Design your model carefully — star schemas perform best, and complex relationships or bidirectional filters can force DirectQuery fallback. Test with representative query patterns before deploying to production.

DirectLake represents a significant step forward for BI workloads — eliminating the freshness vs performance trade-off that has constrained Power BI design for years. For organisations invested in Fabric and OneLake, it's a compelling reason to modernise your analytics architecture.