Fabric Warehouse vs Lakehouse for Gold: when SQL is the product

Medallion debates stall on tools. Bronze and Silver can live in a Lakehouse. Gold is a product decision: who consumes it, and with which contract.

Two contracts

Lakehouse Gold (Delta tables): Spark and Python own the write path. Power BI can sit on Direct Lake. File layout, shortcuts, and notebooks are first-class. T-SQL exists (SQL endpoint) but is not the design center.

Warehouse Gold: T-SQL is the design center. Constraints, views, procedures, and the tools finance already knows. Power BI Import or DirectQuery/Direct Lake-on-warehouse patterns. Analysts with SSMS muscle memory stop opening notebooks to “just add a column.”

Choose Warehouse Gold when

• The consumers are BI authors and finance analysts, not data engineers
• You need views as the stable contract (star schema that outlives a notebook rewrite)
• Row-level governance will be enforced closer to SQL (plus the semantic model)
• Auditors ask “where is the table definition?” and you want a warehouse object, not a notebook cell

Choose Lakehouse Gold when

• Upstream is already Spark-heavy and the BI team is comfortable on Direct Lake over Delta
• Gold is wide files reused by data science and BI, not a classic star
• You are still discovering grain and do not want warehouse DDL churn every sprint

Hybrid that works

Bronze + Silver in Lakehouse. Certified Gold marts in Warehouse, fed by pipelines or shortcuts. Semantic models bind to Warehouse Gold (or Direct Lake on those tables), never to Bronze.

Do not duplicate Gold in both “for safety.” You will get two revenues.

Migration slice

Pick one subject area (e.g. billed revenue). Land Warehouse Gold as views over Silver shortcuts or as loaded tables. Point one semantic model at it. Compare: author time, refresh, query latency, “who broke the view.” Then decide the next domain.

Picking Warehouse vs Lakehouse for a Gold mart that finance will live in? Book a 30-minute call.