GPU-accelerated vector (k-NN) indexing on Amazon OpenSearch Service and OpenSearch Serverless lets you build billion-scale vector indexes in hours instead of days. This post goes deep on the decoupled GPU architecture, the CAGRA-to-HNSW conversion, a one-billion-vector benchmark, and operational best practices for production.