AWS Lambda introduces MicroVMs for isolated code sandboxes

Category: Infrastructure
Status: confirmed
Sources: AWS News Blog, discussion
Summary: AWS announced Lambda MicroVMs on 2026-06-22, a serverless primitive for running untrusted user-generated or AI-generated code with virtual-machine-level isolation. Each MicroVM is powered by Firecracker, runs on ARM64 with up to 16 vCPUs, 32 GB memory, and 32 GB disk, supports up to 8 hours of total runtime, and can suspend after a configurable idle period and resume from a snapshot with memory, disk, and process state intact. AWS lists AI coding assistants, interactive code environments, data analytics, vulnerability scanners, and game servers as target use cases.
Comments: HN commenters compared it to Firecracker directly, E2B, and Google Cloud Run gen2, and questioned whether the 8-hour total-runtime cap rules out long-lived developer environments.
Why it matters: A managed Firecracker sandbox with suspend and resume gives agent and code-execution platforms a first-party isolation primitive instead of a self-built one.
Follow-up: Watch for the pricing detail, GPU support, and whether the 8-hour cap changes for persistent environments.