Gold layer in Databricks and connect to Power BI - this is a good option.

However, If you need to use some of Fabric capabilities, because your team has preferences to use T-SQL, Direct Lake, Python notebooks, low-code tools like Data Factory. MS Fabric as a gold layer may be an option.

You can integrate Databrick into Fabrick using Databricks mirroring, but this feature is in preview, alternatively you can use short cuts, but if you use mange tables this will be more complicated because names of tables aren't predictive, you need to read them from UC.

To give a recommendation I'd discuses it with a client to understand their use cases.

Great points from everyone. In my experience, the choice really hinges on whether you need a unified, metadata-driven access layer or a centralized storage layer for raw and transformed data. That’s essentially the core difference between a data fabric vs data lake approach — fabric focuses on integration and governance across distributed sources, while a lake is optimized for scalable, cost-effective storage and batch/ML workloads.

For teams leaning toward Fabric in the Gold layer, it makes sense if you want tighter T-SQL/Power BI integration and low-code orchestration. If you’re already invested in Databricks and want to minimize duplication, keeping Gold in Databricks and using Serverless SQL for Power BI can be simpler and more cost-efficient.

We’ve explored these trade-offs in more detail in this write-up on data fabric vs data lake, which might help frame the decision based on specific use cases.

When looking at a data fabric vs data lake  strategy, the data lake approach shines when you need a high-performance landing zone for vast amounts of structured, semi-structured, and unstructured data. It gives data science and engineering teams total freedom over raw storage and heavy-duty batch processing. On the other hand, a data fabric acts as an intelligent connective tissue—abstracting the complexity of disparate storage layers so business teams get governed, self-service access without needing to physically centralize every byte.

If you’re structuring your Medallion architecture (Bronze, Silver, Gold), bringing a Data Fabric model into the Gold layer works well for organizations that prioritize real-time semantic modeling, unified access policies, and direct BI integration. However, if your data estate already heavily relies on an open-lakehouse model like Databricks or Delta Lake, pushing your Gold layer directly to fabric storage might introduce unnecessary ETL overhead. In those cases, serving direct queries straight from the lakehouse via lightweight compute engines is often the cleaner, more cost-effective route.