ADF vs Databricks vs Synapse: which data platform?
Verdict: Pure orchestration and movement is Data Factory. Lakehouse and ML - Delta Lake, MLflow, Unity Catalog - is Databricks. T-SQL warehousing plus Spark in one workspace, with PREDICT in SQL, is Synapse.
| Criterion | Azure Data Factory | Azure Databricks | Azure Synapse Analytics |
|---|---|---|---|
| Job | ETL orchestration: 90+ connectors, code-free copy, mapping data flows | Lakehouse and collaborative data science | Unified analytics workspace |
| Native stack | Serverless Spark under mapping data flows - no notebooks | Delta Lake, MLflow, Unity Catalog, notebooks | Dedicated + serverless T-SQL, Spark pools, Data Explorer, embedded ADF engine |
| ML angle | None - orchestrates other services | MLflow experiment tracking first-class | T-SQL PREDICT scores models inside SQL |
| Choose when | Data movement only - no analytics workspace wanted | Complex ML, fine-grained governance via Unity Catalog | T-SQL warehouse + ad-hoc SQL over Parquet + Spark, one workspace |
Rules
- Choose Data Factory when the requirement is orchestration across 90+ connectors and explicitly not an analytics workspace.
- Choose Databricks when Delta Lake, MLflow and Unity Catalog are named - Synapse's governance is Synapse RBAC and MLflow is not native there.
- Choose Synapse for T-SQL warehousing plus serverless SQL over the lake plus Spark in one workspace - and PREDICT for in-SQL model scoring.
Traps
- Synapse embeds the same ADF engine - but picking Synapse for pure data movement pays for a workspace the team will not use.
- Databricks SQL Warehouse is not T-SQL and has no PREDICT - it fails the Synapse scenario on both counts.