[DigitalToday reporter Chi-gyu Hwang (황치규)] AWS has added a microVM feature to its serverless platform Lambda.

According to a recent report by The Register, the microVM is based on AWS' own lightweight virtual machine technology, Firecracker, and extends Lambda's 15-minute execution limit to up to 8 hours.

Developers upload a Dockerfile defining a container and application files to Amazon S3. Lambda converts them into Firecracker snapshots and runs them across multiple instances as needed. AWS presented key uses for microVMs as isolated code execution environments, including scanning potentially malicious packages, vulnerability scanning, and preventing AI-generated code prompt injection and unsafe outputs. It added that they can also be used in CI/CD pipelines.

MicroVMs are also suitable for running AI agents. They are similar to the AgentCore runtime that AWS already provides, but microVMs are more general-purpose and, unlike the AgentCore runtime, can be suspended and resumed.

MicroVMs can have multiple states, including running, suspended and terminated. If there is no traffic, they are automatically suspended, and when new requests come in, they resume while maintaining the existing state. They also support automatic scaling of up to 4 times the baseline specification.

Pricing is charged per second based on virtual CPUs, RAM, snapshot storage and data transfer. Computing costs are not incurred while suspended, which can reduce the cost burden of long-running tasks. It is currently available only in the U.S. East, U.S. West, Tokyo and Ireland regions, and supports only ARM-based AWS Graviton instances.