The five platforms meter the same four layers differently, and the maintenance that Where Costs Come From found neglected in BookNest's lake is exactly what each one now sells:
| Platform | Table storage | Maintenance | Catalog | Query compute |
|---|---|---|---|---|
| Databricks 2,717 | Your cloud bucket | Serverless DBUs | In the platform | DBUs (5.14.1) |
| Snowflake | Your bucket or Snowflake's | Warehouse credits | Included or Open Catalog | Credits (4.13.2) |
| AWS 24 | $0.0265/GB-month | $0.002/1,000 objects + $0.005/GB | Glue: first million free | Athena, EMR, Glue |
| Cloud Storage rates | Data Compute Units | Runtime catalog | On-demand or slots (4.13.3) | |
| Microsoft Fabric 6 | $0.026/GB-month hot | Capacity units | OneLake catalog | $0.18/CU-hour (4.13.6) |
Three patterns stand out. Storage stays cheap and nearly equal: the published rates above sit between $23 and $27 per TB-month, and three of the five leave storage on your own cloud bill. Maintenance moves from your cron jobs to their meter: compaction, snapshot expiry and orphan removal (Iceberg in Production and Security and Lakehouse Costs) become a line item you cannot schedule, only budget, so watch it in the first month. Compute decides the bill, and its units (DBUs, credits, slots, DCUs, CUs) differ so much that only a measured workload compares them: run one week of BookNest's real queries on each trial, as Comparing Pricing Models's model suggests, and multiply out.
Two costs hide outside these columns. Cross-region and cross-cloud transfer applies whenever an engine on one cloud reads tables on another, the very scenario open formats invite. And catalog lock-in is cheaper to avoid on day one than to undo later: prefer a platform whose catalog speaks the Iceberg 129 REST API to outside engines.