One Copy, Two Engines: Architecting a Governed Lakehouse Across Fabric and Azure Databricks

Short summary

Microsoft and Databricks have converged on OneLake and Unity Catalog, enabling bi-directional zero-copy interoperability between Fabric and Azure Databricks as of mid-2026. The article dissects three distinct integration mechanisms — Mirrored Azure Databricks Catalog (GA), OneLake external locations for Unity Catalog (Beta), and Publish to Fabric (Preview) — each with different data-flow semantics. The key architectural decision is where managed tables physically live, driven by who owns the write path.

• •Three integration mechanisms exist: Mirrored Catalog (GA), OneLake external locations (Beta), and Publish to Fabric (Preview) — each with different data-flow semantics
• •Zero-copy bi-directional interoperability is now possible, eliminating the old two-copy mess between Databricks and Fabric
• •The critical design choice is where managed tables physically live, determined by which platform owns the write path

Generated with AI, which can make mistakes.