AWS rebuilds AgentCore Runtime with elastic memory and ~2-second cold starts

AWS has released a new generation of AgentCore Runtime, the serverless microVM compute layer inside Amazon Bedrock AgentCore for running AI agents. The headline change is elastic memory: sessions start small, grow on demand and hand unused memory back while they run, instead of holding a fixed allocation. Cold starts are now snapshot-based, so startup time stays consistent regardless of container image size. AWS cites a P75 cold start of roughly 1.9 to 2.0 seconds, against 5.4 to 30 seconds on the first version. The service keeps its serverless model: no pre-provisioning, scale to zero, and hardware-enforced isolation per session. It is available in us-east-1, us-east-2, us-west-2, eu-west-1 and ap-northeast-1. For teams running agent workloads on AWS, the practical takeaway is lower idle cost and more predictable latency for bursty, interactive agents; image-heavy agents benefit most.

This is a summary by our content curator. Read the original at AWS What's New: https://aws.amazon.com/about-aws/whats-new/2026/09/new-agentcore-runtime-generally-available/.