For a Fabric workspace with a large lakehouse containing historical data and frequent updates, which semantic model storage mode yields best performance for near real-time reporting?

• A. Direct Lake
• B. Import and Direct Lake combined
• C. DirectQuery
• D. Import

Answer: (A)

Direct Lake uses the lakehouse as the primary source of truth and reads data directly from the lake with an optimized query path, avoiding the delays of importing and refreshing data. With frequent updates in a large lakehouse, this mode keeps reports current because queries pull the latest data from the storage layer instead of relying on periodic import cycles. The semantic model can benefit from fast table/partition pruning and metadata that speeds up access, making near real-time reporting more responsive as data changes.

In contrast, importing copies data into the model, so updates in the lake aren’t visible until a refresh occurs—adding latency and management overhead for near real-time needs. DirectQuery is also live, but it often incurs higher latency across the network and the source system, especially with very large datasets, which can hinder responsiveness. The combined Import and Direct Lake approach adds complexity and still won’t beat Direct Lake for fresh, up-to-the-minute results in a large, frequently updated lakehouse.