Possibility to Import Data from Semantic Model Without Triggering XMLA Read Operation

Thank you for the detailed question.

A quick clarification: XMLA read operations are not inherently the cause of capacity issues. They are the mechanism the Analysis Services engine uses to serve semantic model queries. The actual capacity impact depends on model size, query complexity, concurrency, and your capacity SKU.

If you are reading data directly from a semantic model through the Analysis Services connector, "Connect to semantic model" in Power BI Desktop, Analyze in Excel, DAX Studio, or the REST executeQueries API the request is ultimately served by the Analysis Services engine, so some query workload against the model is unavoidable.

To reuse the final numbers while minimizing load on the semantic model, the common patterns are:

1. Read from the upstream source (Lakehouse, Warehouse, SQL, source system) instead of the semantic model.

2. Materialize the required outputs into a Lakehouse/Warehouse/dataflow table as part of your existing ETL, and consume that table downstream.

3. Use Dataflows (Gen1 or Gen2) built on the same source to produce reusable output tables.

Before redesigning, use the Microsoft Fabric Capacity Metrics app to confirm whether XMLA/interactive queries are actually the main CU driver, versus refreshes or other background operations.

If the logic exists only as DAX measures in the semantic model, you'll either need to recreate it upstream, or keep querying the model and focus on optimizing the model/DAX and controlling query frequency.

References:

• Semantic model connectivity (XMLA endpoint): https://learn.microsoft.com/fabric/enterprise/powerbi/service-premium-connect-tools

• Monitor Fabric capacity: https://learn.microsoft.com/fabric/enterprise/metrics-app

• Dataflows overview: https://learn.microsoft.com/power-bi/transform-model/dataflows/dataflows-introduction-self-service

Thanks,

C Srikanth

Community Support Team