GPU acceleration for vector indexing is now available on Amazon OpenSearch Service and Amazon OpenSearch Serverless, enabling scaling to billions of vectors. Powered by NVIDIA cuVS, this capability offloads compute-intensive vector index building to specialized GPU workers while existing CPU infrastructure continues serving search, resulting in faster, more cost-efficient construction of large-scale vector indexes without sacrificing query performance. The post details the decoupled architecture, explains how a GPU-built index is converted into one that CPU data nodes can search with no quality tradeoff, and includes a benchmark of one billion 1024-dimensional vectors. Use cases include adopting new embedding models faster, where GPU acceleration shortens a CPU rebuild that could take days or weeks to hours; accelerating large-scale reindexing for a global ecommerce application managing billions of product listings; absorbing bursty or high-sustained writes for a media company covering events like the World Cup or Olympics, which needs to index millions of real-time embeddings simultaneously; and right-sizing clusters for mixed read/write workloads, where a retail system traditionally over-provisions its CPU cluster to handle peak indexing loads during catalog refresh.