| Engine | Licence | Runs as | Best at | Weak at |
|---|---|---|---|---|
| PostgreSQL 18 1,289 | PostgreSQL License (open source) | Server | Transactions, point lookups, RLS | Scans beyond one node |
| DuckDB 1.5 61,228 | MIT | Library in your process | Files and local analysis | Many concurrent writers |
| ClickHouse 26.9 29,491 | Apache 2.0 | Server or cluster | Event streams, many users | Updates, point lookups |
| Snowflake | Proprietary service | Managed, multi-cloud | Elastic concurrency, sharing | Cost control, lock-in |
| BigQuery 1 | Proprietary service | Serverless, Google Cloud 1 | Zero administration | Scan-priced surprises |
| Redshift 24 | Proprietary service | Managed, AWS 24 | AWS integration | Tuning keys and sizes |
| Databricks 2,717 SQL | Proprietary on open formats | Managed, multi-cloud | Lakehouse, Spark 129 and SQL | Platform complexity |
| Fabric Warehouse | Proprietary service | SaaS capacity, Azure 6 | Power BI, OneLake | Capacity sizing |
The open-source engines cost nothing to license but everything to operate: upgrades, backups, monitoring and capacity are yours. All three also exist as managed services (many PostgreSQL providers; MotherDuck for DuckDB; ClickHouse Cloud), so "open source" decides who may run the engine, not who must. The transformation layer is open in every case: dbt Core 37,942 , SQLMesh and MetricFlow are Apache 2.0, while dbt v2's full distribution adds a proprietary licence on top of an Apache 2.0 core (The dbt Fusion Engine).