Apache Iceberg pricing guide details
Pricing Overview
Apache Iceberg is free and open source under Apache 2.0 — no license cost for the table format itself. Real costs come from three components you compose around Iceberg: query engines (Spark, Trino, Flink, or managed services), object storage (S3, GCS, Azure Blob), and catalog infrastructure (AWS Glue Data Catalog, Hive Metastore, or an Iceberg REST catalog). This dual model is Iceberg's defining pricing advantage: there's no per-table, per-query, or per-user license fee. Commercial managed Iceberg services exist (Tabular, now part of Databricks; Dremio; cloud vendor offerings) for teams that want managed deployments, but the core format is free forever.
For large enterprises running petabyte-scale lakehouses with multiple query engines, the bill is driven by query volume and storage scale, not Iceberg licensing. The format itself is infrastructure, not a product with tiered pricing.
Plan Comparison
Iceberg doesn't have named plans — you compose costs from underlying components:
| Component | Pricing | Notes |
|---|---|---|
| Iceberg format | Free (Apache 2.0) | The core spec and Java/Python reference implementations |
| Object storage | Billed per GB-month on S3, GCS or Azure | Dominant cost for large tables |
| Query engine (self-hosted) | Free (Spark, Trino, Flink) | Infrastructure + ops time |
| Query engine (AWS Athena) | Billed per TB scanned | Pay-per-query; good for ad-hoc |
| Query engine (Snowflake) | Snowflake credits (Standard from $2/credit, Enterprise higher) | Reads Iceberg natively |
| Query engine (Databricks) | DBU-based (varies by SKU) | Native Iceberg support through Unity Catalog |
| AWS Glue Data Catalog | $1 per 100K requests, $1/month per 100K objects stored | Most common managed catalog |
| Iceberg REST catalog | Free (self-hosted) or managed via Tabular/Dremio | Modern catalog standard |
Hidden Costs and Considerations
Three cost drivers catch teams off-guard:
- Metadata operations scale with table size — a table with millions of partitions generates millions of metadata files. AWS Glue Data Catalog charges per request, and poorly tuned queries can generate surprising metadata bills.
- Small-file problem from streaming ingestion — Flink or Spark Structured Streaming writes create many small Parquet files. Without scheduled compaction, query performance degrades and storage costs rise. Compaction jobs themselves cost compute time.
- Orphan file cleanup is manual — Iceberg snapshots accumulate forever unless you run
expire_snapshotsandremove_orphan_files. Left unchecked, these consume 10-100x the active table size.
Volume discounts exist on S3 (savings above 500 TB/month) and on managed query engines (Databricks and Snowflake both offer committed-use contracts). Tabular and Dremio commercial services have vendor-specific pricing for managed Iceberg deployments.
Cost Estimates by Team Size
Iceberg itself is free, so there is no licence cost to estimate. What you pay is your query engine and your object storage, and both are billed by the vendor you choose at rates they publish. What changes with team size is which components you run:
- Solo engineer / proof-of-concept: S3 plus Athena for ad-hoc queries, billed per TB scanned, with an AWS Glue catalog.
- Small team (5 engineers, <1 TB active data): the same, or a small Trino cluster if query volume makes per-query billing worse than a running cluster.
- Mid-size team (20 engineers, 10-100 TB data): a managed Spark platform such as EMR, Databricks or managed Flink, a larger S3 footprint, and enough query volume that engine cost dominates storage.
- Large enterprise (100+ engineers, petabyte scale): multiple engines against one table format, committed-use contracts, and object storage large enough to negotiate on.
Price each of those against the engine's own rate card. Most teams underestimate the query engine and overestimate the storage.
Most teams underestimate query-engine costs — the format is free but the engines reading it aren't. Budget 60-80% of total Iceberg-based spend on query engines, 15-30% on storage, 5-10% on catalog and operations.
How Apache Iceberg Pricing Compares
Iceberg's free-format model differs fundamentally from proprietary warehouses:
- Snowflake: Credit-based pricing. Zero operational overhead but typically 5-10x the cost of Iceberg plus Spark or Trino at scale. Worth the premium for teams without data platform engineering capacity.
- Databricks: DBU-based pricing, supports Iceberg natively. Choose Databricks when you want a managed lakehouse and will commit to the platform; Iceberg there is a feature rather than the foundation.
- Google BigQuery: Serverless warehouse with native Iceberg support and BigLake managed tables. Storage is billed per GB-month, queries $6.25/TB scanned. Pricing is competitive; choose BigQuery on GCP.
- Delta Lake: Also free (Apache 2.0), similar cost structure. Choose Delta when committed to Databricks; Iceberg wins on multi-engine flexibility.
- Apache Hudi: Also free, similar cost structure. Choose Hudi for streaming-upsert-heavy workloads; Iceberg wins on analytics query performance.
The honest summary: Iceberg is cheapest for teams willing to assemble the components themselves. For teams valuing operational simplicity over cost, proprietary warehouses or managed lakehouse platforms (Databricks, Dremio) are the path of least resistance. The Iceberg format itself never costs money — only the surrounding infrastructure does.