Choosing the right storage mode for Power BI reports fed by Azure Databricks can often feel like a roll of the dice. Developers and architects frequently rely on intuition rather than data, leading to potential performance bottlenecks that frustrate end users.

A recent white paper aims to change this by offering concrete benchmarks on four different storage modes, helping to illuminate the best choices for various reporting scenarios.

The study compares the following options for serving Delta tables to Power BI reports:

1. Direct Lake on OneLake — Utilizing a Fabric lakehouse or warehouse.

2. Direct Lake on mirrored Unity Catalog tables — Implementing shortcuts without duplicating data.

3. DirectQuery — Accessing data through a Databricks SQL warehouse.

4. Composite Model on Databricks — Combining DirectQuery with Import-mode aggregations.

The findings reveal that while there’s no single best choice, clear patterns emerge based on data volume and query types. For smaller and mid-sized datasets, Direct Lake on OneLake stands out, delivering fast, interactive performance without requiring additional modeling effort.

This could mean that a manager reviewing regular reports might now find them loading quicker and displaying data more responsively, easing decision-making processes.

However, when dealing with vast datasets—think billions of rows—the results shift. The Composite Model with aggregations proves to be the fastest, maintaining sub-100ms query response times for aggregated data. But there’s a catch: if queries need to pull from the underlying DirectQuery source, performance can lag significantly.

This means that while the model can handle many common queries efficiently, it’s critical to ensure that the aggregations align with actual user behavior to avoid pitfalls.

It’s essential to approach these findings with caution. The benchmarks are a snapshot from mid-2026 and both Power BI and Azure Databricks are rapidly evolving. Moreover, the study focuses on the end-user experience within Power BI rather than the raw execution speed of the database itself. Thus, organizations should consider conducting their own tests tailored to their unique data and use cases.

For developers and data architects, the white paper serves as a valuable resource to guide their architecture decisions. It delves into the specifics of the test setup, measurement criteria, and how results vary based on different variables like cache state and filter scenarios.

As organizations increasingly rely on data-driven insights, understanding the nuances of these storage modes could lead to more efficient reporting and happier end users.

In an era where user experience can make or break a reporting tool, the insights from this benchmarking study offer a much-needed data-driven foundation for making architectural decisions. The full white paper is available for those looking to dive deeper into the findings and implications for their Power BI implementations.

For more information, check out the white paper here and the Power BI download center.