Evolution: S3 Tables and Vectors

How AWS S3 is built

Topic 9 of 13

Published Jan 21, 2026 · Clip 49:07–1:01:16 (12:09)

The guest traces the rise of parquet (~2020) and Iceberg as a replacement for Hive, predicts the world's tabular data will live in S3, and cites Supabase writing directly into S3 Tables. SQL becomes a native property of S3 via tables—the lingua franca of data that LLMs know well. She explains S3 Vectors as a new primitive for embeddings, how precomputed vector neighborhoods enable sub-100ms warm queries, up to 2 billion vectors per index and 20 trillion per vector bucket, all designed so scale works to S3's advantage.