This article announces a ground-up re-architecture of Amazon OpenSearch Serverless, delivering significant performance and cost improvements for AI agents and dynamic workloads.

• Autoscaling 20x faster with compute provisioning in seconds instead of minutes
• Scale to zero capability eliminates compute costs during idle periods (10-minute timeout)
• Up to 60% lower costs compared to provisioning for peak capacity
• Decoupled compute and storage enables independent scaling of indexing and search
• GPU acceleration for vector index construction automatically reduces indexing time
• Express Create simplifies collection setup with no upfront configuration needed
• New static regional endpoint serves all collections, improving multi-tenant workload management
• Collection Groups required for new architecture, enabling shared compute across collections
• NextGen is default for new collections; Classic architecture still available
• Includes hands-on tutorial for creating vector collections and observing scale-to-zero behavior

The NextGen architecture makes OpenSearch Serverless ideal for agentic AI workloads requiring dynamic, unpredictable scaling with minimal operational overhead.