Connecting Snowflake to Fabric + Power BI

[AntoineW]

Hi everyone,

I’ve been working primarily with Power BI for years — mainly as a semantic and visualization layer on top of enterprise data sources like Snowflake.

Recently, we migrate to Microsoft Fabric to go beyond classic Power BI capabilities.

My goal is to connect Fabric and Power BI to Snowflake optimally, to leverage all these Fabric capabilities without moving or duplicating data from Snowflake into OneLake.

So far, here’s what I’ve found and experimented with:

• DirectQuery mode and mirroring Snowflake offer two strong patterns for live data access.

• The Fabric semantic model helps standardize metrics while applying governance and RBAC centrally.

• Purview can catalog and classify Snowflake datasets through Fabric integration.

• AI/Notebook integration (Semantic Link, Spark, Python) allows advanced analytics without breaking lineage.

I’d love to hear from others who’ve worked on similar hybrid setups:

• What’s the best design pattern you’ve seen for connecting Fabric + Power BI to Snowflake, keeping minimal or no data duplication?

• How do you handle cross-platform monitoring (Fabric capacity, Power BI usage, Snowflake compute costs)?

• Any best practices for governance and synchronization between Snowflake transformations and the Fabric semantic model?

• Have you discovered any new or upcoming features (e.g., Mirroring improvements, lineage sync, metadata propagation) that make this integration smoother?

Thanks a lot for your insights and experience sharing!

[Vinodh247 · Accepted solution]

You are right with DirectQuery and mirroring as the 2 main integration options between Snowflake, Fabric, and PBI. 

Below are my view:

For minimal duplication, use DirectQuery from fabric’s semantic model to snowflake. This keeps snowflake as the compute layer while enabling centralized governance, metric definitions and security in fabric. Mirroring becomes useful if you need consistent query perf or integration with fabric pipelines, notebooks or AI workloads but  note that it involves data replication into onelake.

For cross platform monitoring, combine fabric capacity metrics (activity logs, capacity metrics app) with snowflake resource monitors and PBI’s admin APIs to track usage and cost holistically.

For governance, align fabric semantic models with snowflake roles and policies. Use purview for unified lineage and classification across both.

Upcoming fabric features like enhanced snowflake mirroring, lineage synchronization and metadata propagation into Purview are expected to simplify this hybrid design.

Overall, a DirectQuery based semantic layer on top of snowflake, enriched with fabric AI and governance capabilities, is currently the best low duplication pattern.

[v-sshirivolu · Accepted solution]

Hi @AntoineW ,

I would also take a moment to thank @Vinodh247 , for actively participating in the community forum and for the solutions you’ve been sharing in the community forum. Your contributions make a real difference.

Fully agree with your summary on DirectQuery and mirroring as the core integration patterns between Snowflake, Fabric, and Power BI. A few additional points that might further strengthen this approach - leveraging Snowflake materialized views for frequently queried data can optimize performance and cost while keeping duplication minimal. It also helps to ensure Fabric semantic model fields align closely with Snowflake schema definitions to maintain consistent metrics and lineage. Including Power BI dataset and report usage metrics alongside Fabric capacity and Snowflake resource monitors provides a more complete cross-platform monitoring view. Additionally, using Purview integration for automatic metadata and schema propagation keeps transformations synchronized. For advanced analytics, Fabric Semantic Link and Notebook integration allow AI and Python workloads to operate directly on live Snowflake data without breaking lineage. 
I wanted to check if you had the opportunity to review the information provided. Please feel free to contact us if you have any further questions.

[Vinodh247 · Accepted solution]

You are right with DirectQuery and mirroring as the 2 main integration options between Snowflake, Fabric, and PBI. 

Below are my view:

For minimal duplication, use DirectQuery from fabric’s semantic model to snowflake. This keeps snowflake as the compute layer while enabling centralized governance, metric definitions and security in fabric. Mirroring becomes useful if you need consistent query perf or integration with fabric pipelines, notebooks or AI workloads but  note that it involves data replication into onelake.

For cross platform monitoring, combine fabric capacity metrics (activity logs, capacity metrics app) with snowflake resource monitors and PBI’s admin APIs to track usage and cost holistically.

For governance, align fabric semantic models with snowflake roles and policies. Use purview for unified lineage and classification across both.

Upcoming fabric features like enhanced snowflake mirroring, lineage synchronization and metadata propagation into Purview are expected to simplify this hybrid design.

Overall, a DirectQuery based semantic layer on top of snowflake, enriched with fabric AI and governance capabilities, is currently the best low duplication pattern.