AWS has debuted Lambda MicroVMs, a new serverless computing capability built to run user-supplied and AI-generated code inside virtual machine environments without requiring developers to manage underlying virtualization infrastructure.
The launch addresses a challenge facing software vendors building AI-powered applications. As coding assistants and developer tools generate code automatically, companies need secure ways to execute code that may be vulnerable or even malicious. Lambda MicroVMs provide each user, session or workload with its own isolated execution environment.
“VM-level isolation just became a native serverless primitive,” Mitch Ashley, VP of Software Lifecycle Engineering at The Futurum Group, told Techstrong.it. “Per-session hardware isolation that is stateful, resumable, and billed only while running targets a specific problem: safely executing code the developer never wrote and cannot fully trust.”
Furthermore he added, “For teams running agents and AI-generated code, the containment boundary sets the ceiling on autonomy, and cheap resumable isolation raises it. The hard part remains: a sandboxed VM still reaches the network, so the boundary holds only as well as the egress controls around it.”
Built on Firecracker
The new offering is built on Firecracker, AWS’s lightweight virtualization technology that already underpins the company’s Lambda service. Firecracker currently supports more than 15 trillion Lambda executions per month, giving AWS a foundation for expanding serverless workloads into VM-based environments.
Unlike standard Lambda functions, which have historically been limited to 15-minute execution windows, Lambda MicroVMs can remain active for as long as eight hours. The extended runtime makes the service suitable for workloads that require sustained processing, including vulnerability analysis, package inspection, continuous integration pipelines and AI agent operations.
Developers create MicroVM images using Dockerfiles and application artifacts. AWS converts these images into Firecracker snapshots that can be launched repeatedly as needed. Each MicroVM receives its own HTTPS endpoint and supports protocols including HTTP/2, gRPC and WebSockets.
To support state preservation, MicroVMs can be suspended when inactive and later resumed without losing application context. This allows workloads to retain memory and execution state while reducing compute costs during idle periods.
Security Benefit of Isolated Environments
As companies deploy AI systems that generate code autonomously, security is a growing concern. Running AI-created software in isolated environments reduces the risk that flawed code affects other applications or infrastructure.
The new capability also creates another option for building AI agents. AWS already offers AgentCore Runtime for agent execution, but Lambda MicroVMs provide a more general-purpose environment while adding the ability to suspend and resume workloads.
Additionally, because MicroVMs provide shell access, networking flexibility and serverless consumption-based pricing, they support teams that want lightweight environments for development and long-running application processes.
Pricing is based on resource consumption, including virtual CPUs, memory, storage snapshots and data transfer. Compute charges stop when a MicroVM is suspended, leaving customers responsible primarily for stored state until execution resumes.
At launch, Lambda MicroVMs are available in AWS regions including Northern Virginia, Ohio, Oregon, Tokyo and Ireland. Current support is limited to Arm-based AWS Graviton infrastructure.
