Compare
LakehouseBox and Amazon S3 Tables
Both keep Apache Iceberg tables behind a managed catalog. S3 Tables is part of AWS; LakehouseBox is a small European service that an agent can start without an account.
Side by side
Their column is what their own pages said on 12 October 2026 (sources below); prices are theirs, in their currency, before tax.
| Amazon S3 Tables | LakehouseBox | |
|---|---|---|
| Who operates it, where | Amazon Web Services (a US company); table buckets in eight European regions, Frankfurt among them. The AWS European Sovereign Cloud does not offer S3 Tables | Lakehousebox, S.L. (Spain, in formation); tables and files stored at Hetzner in Nuremberg, Germany; a locked backup copy in France |
| To start | An AWS account | An agent needs no account: lhbox new gives it a catalog it can write for 72 hours, until a person claims it. A person: a free account, 5 GB, no card |
| Price | Frankfurt: $0.0282 per GB-month for the first 50 TB; $0.0054 per 1,000 writes and lists, $0.00043 per 1,000 reads; monitoring $0.025 per 1,000 objects a month; compaction $0.00215 per 1,000 objects plus $0.005 per GB | Free up to 5 GB. Only storage is metered: no request, egress or compute fees. Paid storage is not open yet (announced at €0.10 per GB-month) |
| Catalog and credentials | Iceberg REST endpoint signed with AWS SigV4 (no OAuth); credentials vended through the Glue endpoint once Lake Formation grants are set | Iceberg REST catalog; each engine gets storage credentials for one table at a time, valid for an hour |
| Engines | Spark, EMR, Glue, Athena, Redshift; DuckDB attaches a table bucket by ARN | DuckDB 1.5.5+, PyIceberg 0.12, Spark 3.5 (status per engine) |
| Iceberg v3 | Yes, including geometry and geography; Athena cannot read v3 tables | Yes: DuckDB 1.5.5 writes format-version 3, geometry columns included |
| Maintenance | Compaction and snapshot management, billed as above | Compaction, snapshot expiry and orphan-file cleanup every 6 hours, included |
| Compute | Yours, or AWS's analytics services | Yours: your agent or laptop runs the engine. Server-side read-only SQL (20 s per query) through the API and MCP |
| MCP | Not part of S3 Tables | Remote server, read and write (one SQL statement per call) |
| SLA | Yes, AWS's | None yet; one site, no standby |
Which one
Choose Amazon S3 Tables when you already run on AWS and want Spark, EMR, Glue or Redshift on the same tables; you need petabyte quotas, AWS's compliance programme and IAM or Lake Formation governance; and cents per GB matter more than who operates it.
Choose LakehouseBox when your agent should be able to start on its own, without a cloud account or IAM policies; you want one meter (storage) and no per-request or monitoring fees; or you want the operator and its storage provider to be European companies. On the CLOUD Act, AWS's own page says it "applies to all … providers that operate or have a legal presence in the U.S."
Start with LakehouseBox: let your agent start without an account, or sign up free. Other comparisons: all of them.
Sources
Checked 12 October 2026. Something out of date? Write to hello@lakehousebox.com and we correct it.
- AWS price list for S3 in eu-central-1
- S3 Tables with open-source engines (endpoint, SigV4)
- S3 Tables through the Glue Iceberg REST endpoint
- S3 Tables and Iceberg v3
- AWS European Sovereign Cloud: S3
- AWS on the CLOUD Act
- DuckDB: Iceberg catalogs
Ours: pricing, sovereignty, engines, capabilities.