S3 Vectors goes live: 2B vectors per index, 100ms queries

S3 Vectors just hit general availability and it's a serious upgrade. We're talking 2 billion vectors per index now—that's 40x what preview offered. Query latency sitting at 100ms or less for frequent searches. Write throughput at 1,000 PUT transactions per second. Serverless means zero infrastructure to manage.

The cost story is wild too. Up to 90% savings versus specialized vector databases. No sharding headaches. No complex federation logic. Just throw your entire vector dataset into one index and let S3 handle it.

What's live right now: Bedrock Knowledge Base integration for RAG apps. OpenSearch integration for search and analytics on top of your vectors. CloudFormation support for infrastructure as code. PrivateLink for private network access. 14 AWS regions (expanded from 5 during preview).

The performance improvements matter if you're building conversational AI, multi-agent systems, semantic search, or RAG applications. Infrequent queries under 1 second. Frequent queries around 100ms. You can pull 100 search results per query now instead of 30. That's more context for your RAG pipelines.

The adoption numbers during preview were insane: 250,000 vector indexes created. 40 billion vectors ingested. Over 1 billion queries in just four months. People were hungry for this.

Pricing is straightforward. You pay for what you ingest (PUT pricing per logical GB), what you store, and what you query. Beyond 100,000 vectors, the per-TB query cost drops. No surprises.

If you're managing vectors across multiple specialized databases, consolidating into S3 Vectors could cut your bill significantly. The serverless model means you're not paying for idle capacity. Just vectors in, vectors out, and the queries you run.