How would you architect a data lake on AWS that serves both analytics and ML training?

Updated Aug 1, 2026

Short answer

Use S3 with bronze/silver/gold zones, an open table format such as Iceberg for ACID and time travel, Glue Catalog for metadata, Lake Formation for fine-grained governance, and expose curated gold tables to both Athena for analytics and SageMaker for training.

Deep explanation

Analytics and ML have different access patterns — analytics scans aggregates, ML reads full feature rows repeatedly — but they must share one governed source, or the numbers stop agreeing.

Zone model

Bronze — raw, immutable, exactly as ingested. Partition by ingestion date. Never mutated; it is your ability to reprocess history.
Silver — cleaned, deduplicated, conformed types and schema.
Gold — business-level aggregates and ML-ready feature tables.…

Unlock with a Pro subscription to view this section.

View pricing