Fabric Lakehouse or Warehouse for the gold layer? Decide from the workload

A team moving into Microsoft Fabric proposes Lakehouse for bronze and silver, then Warehouse for gold because its users know SQL. A practitioner discussion raises a useful challenge: SQL users can also query Delta tables through the Lakehouse SQL analytics endpoint. The deciding factor is what the team must build and operate, not the familiarity of the query language.

Separate engineering writes from consumer reads

Map each gold dataset: who writes it, how often it changes, how it is secured, and whether users need only SELECT or also T-SQL modeling and write operations. Microsoft compares Warehouse and Lakehouse as distinct development experiences over Delta data. The Lakehouse SQL analytics endpoint is a read-only T-SQL surface. Warehouse offers fuller T-SQL write and warehouse development capabilities.

That distinction matters. If Spark or notebooks own all transformations and the audience only queries curated Delta tables, Lakehouse plus its SQL endpoint may suffice. If the gold layer is authored in SQL with transactions, procedures or warehouse tooling, Warehouse may be the cleaner ownership boundary.

Price the whole operating path

Run a small representative workload in both designs. Include ingest, transformation, semantic model refresh, interactive queries and concurrency. Record capacity consumption, latency and deployment friction. Avoid presenting a general performance verdict from one team's Reddit experience; Fabric capacity behavior depends on the actual workload and configuration.

Check the security and lifecycle path too: workspace boundaries, roles, CI/CD, data contracts and how a changed column reaches a dashboard. A design that saves one transformation step but requires manual fixes in every report is not cheaper in practice.

Write a decision record

The handoff should name the chosen item for each layer, the owner of gold definitions, the supported query interface, a capacity budget and a migration route if usage changes. A Lakehouse and Warehouse can coexist, but duplicating every gold table in both without a serving reason creates two places for the same metric to drift.

Adovo’s view

Pick the gold layer where its truth can be owned. The storage label matters less than the write contract, access path and cost of change. If the team cannot say who changes a gold measure and how consumers learn about it, choosing a Warehouse will not make the data trustworthy.