Data Lake

A Data Lake is a large storage repository that keeps raw data in its original format — logs, CSVs, images, audio, database dumps — so the structure can be decided later, when someone knows what question to ask. This inverts the warehouse model, which requires a schema before loading. Storing first and modeling later is cheap because lakes sit on object storage such as Amazon S3, Google Cloud Storage, or Cloudflare R2, where capacity costs far less than warehouse compute. Open table formats then add warehouse-like behavior on top: Apache Iceberg, Delta Lake, and Apache Hudi bring schema evolution, time travel, and transactional writes to files in a bucket, and engines like Databricks, Snowflake, Trino, and DuckDB read them directly. AI teams lean on lakes because model training needs unfiltered text, image, and event data that a curated warehouse table has already thrown away. The pitfall is the data swamp: without catalogs, ownership, and retention rules, a lake becomes petabytes nobody can find or trust. The lakehouse pattern is the current answer, keeping the lake's cheap raw storage while adding the governance and query discipline of a warehouse.