AWS Lambda MicroVMs Extend Serverless Runtimes to 8 Hours with Firecracker Isolation
AWS launched Lambda MicroVMs built on its Firecracker hypervisor, letting developers run isolated Linux containers for up to 8 hours — a massive jump from the existing 15-minute Lambda cap. The service targets AI-generated code execution, long-running CI/CD, and vulnerability scanning.
AWS launched Lambda MicroVMs on June 23, extending the serverless platform’s maximum runtime from 15 minutes to 8 hours. The service runs isolated Linux containers using AWS Firecracker, the open-source virtual machine monitor that already powers Lambda and Fargate under the hood — now surfaced directly to developers as a first-class primitive.
The workflow is straightforward: upload a Dockerfile to S3, and AWS builds and manages the Firecracker snapshot. Developers get full shell access to the container, persistent state via suspension and resume, and auto-scaling that can push the instance up to 4x its base specification when workload demands it. Billing is consumption-based and pauses when the MicroVM is suspended — no idle charges while a job waits for an external event.
The 8-hour cap is not arbitrary. It targets a cluster of workloads that have historically forced teams off Lambda and onto longer-lived compute like EC2 or ECS: long-running CI/CD pipelines, multi-step data processing jobs, and — the primary stated use case in the announcement — safe execution of AI-generated code.
That last use case is the most forward-looking. As AI coding agents produce more code that gets executed in automated pipelines, running that code in an isolated Firecracker VM provides hardware-level containment. If an LLM produces code that attempts to access the host network, read credentials from the environment, or execute a prompt injection payload, the MicroVM boundary catches it. Lambda’s existing isolation is solid; the MicroVM layer adds a dedicated process namespace and tighter resource accounting.
Lambda MicroVMs are launching in select regions — US East (N. Virginia), US West (Oregon), Tokyo, and Ireland — on AWS Graviton (Arm) instances only. Intel/AMD variants are not yet available. The Graviton-only restriction is partly technical (Firecracker’s production-hardened path runs on Arm at AWS scale) and partly a cost signal — Graviton instances are 20% cheaper than equivalent x86 at the same performance tier.
For teams already running Lambda at the edge of the 15-minute cap — using workarounds like chained functions, Step Functions state machines, or fallbacks to ECS — this removes the architectural workaround. The tradeoff is the preview footprint: four regions, Arm only, and no Windows container support at launch.
AWS says broader region availability and x86 support are on the roadmap for later in 2026. Pricing details are in the Lambda pricing page under the MicroVMs section.