AWS Lambda MicroVMs extend runtime to 8 hours

AWS has expanded its serverless computing capabilities with the introduction of Lambda MicroVMs, a feature designed to run isolated Linux containers for up to eight hours. This development marks a significant shift from the 15-minute execution limit previously imposed on Lambda functions, offering developers greater flexibility for long-running tasks while maintaining the benefits of serverless architecture.

The new MicroVMs are built on AWS’s Firecracker virtual machine monitor, which underpins the company’s serverless infrastructure. Developers can define a container using a Dockerfile, upload it to Amazon S3, and let Lambda build it into a Firecracker snapshot. These snapshots can then be deployed as MicroVM instances, which AWS describes as ideal for tasks requiring isolation, such as scanning potentially malicious packages, running AI-generated code, or integrating into CI/CD pipelines.

How it works

Lambda MicroVMs operate as ephemeral, serverless environments that can scale dynamically. Each MicroVM can exist in multiple states, including running, suspended, or terminated. When idle, a MicroVM can automatically suspend, halting compute charges while preserving its state. Upon resumption—triggered by a new network request—the environment restores its previous state, allowing seamless continuation of tasks. AWS has also designed MicroVMs to scale up to four times their base compute specification when demand increases.

Pricing for MicroVMs is consumption-based, calculated per second for vCPU, RAM (provisioned at a 2:1 GB ratio to vCPU), snapshot storage, and data transfer. While compute charges pause during suspension, storage and snapshot costs persist. Currently, the feature is available only in select regions—US East, US West, Tokyo, and Ireland—and supports only Arm-based AWS Graviton instances.

Use cases and limitations

AWS positions MicroVMs primarily as a tool for secure code execution, particularly for isolating untrusted or AI-generated workloads. However, developers have already identified broader applications, including AI agents, extended CI/CD jobs, and even full-day coding sessions. The feature’s ability to provide full shell access and unopinionated HTTP ingress further expands its utility beyond traditional Lambda functions.

"This lets you run anything you like and for up to 8 hours all while giving you full shell access to the VM and unopinionated HTTP ingress. All while still getting the benefits of SnapStart and true consumption-based pricing as opposed to wall-time pricing. It's a complement to functions." — Developer quoted in The Register

Despite these advantages, MicroVMs are not a universal replacement for existing AWS offerings. For instance, the company’s AgentCore Runtime, which also supports eight-hour execution, remains a specialized alternative for AI agents. MicroVMs, however, offer greater flexibility, including the ability to suspend and resume workloads—a feature absent in AgentCore. AWS has even provided a skill for its Agent Toolkit that leverages MicroVMs, simplifying adoption for developers already using its AI agent ecosystem.

What’s next

The introduction of MicroVMs addresses a long-standing request from developers for longer execution times in Lambda. While the feature is currently limited to specific regions and Arm-based instances, its serverless nature and consumption-based pricing model make it an attractive option for workloads requiring both isolation and extended runtime. As adoption grows, AWS may expand regional availability and instance support, further integrating MicroVMs into its broader serverless and AI tooling ecosystem.