Can Lambda MicroVMs work as self-hosted GitHub Actions runners?

Yes, the basic design is feasible. A webhook can trigger the AWS side when a job needs a runner, then start a MicroVM with just-in-time runner configuration. You can use one VM per job rather than keeping a runner fleet alive.

The main fit check is architecture. MicroVMs are arm64 only, so check your build tools, container images, dependencies, and the architecture of the artifacts you need to produce. Hardware configuration is also less flexible than with some other compute options.

Don't compare compute prices alone. A fresh VM for every job can mean pulling images and rebuilding Docker layers repeatedly unless you arrange some external caching. Artifact reuse and networking costs can change the calculation too.

Spot capacity on EC2, or an existing ECS or Kubernetes cluster, may be cheaper for the compute. That's a less useful comparison if you'd have to build and maintain a whole cluster just for runners. Spot interruptions and any warm capacity you keep around also belong in the calculation.

Take a look at CodeBuild hosted GitHub runners before building the webhook and runner provisioning yourself. It can scale to zero and offers different compute sizes, on-demand or provisioned fleets, and custom container options. It may be a simpler fit if the goal is low-maintenance AWS runners rather than specifically trying MicroVMs.

The advertised limits and per-minute rates make MicroVMs interesting, especially for short jobs. The comparison in the discussion used roughly $0.0042 per minute for MicroVMs versus $0.005 for a two-vCPU GitHub-hosted runner, and cited an eight-hour versus six-hour runtime limit. Treat those as starting points, not a quote for your setup. Check current pricing, billing increments, and limits for the configurations you plan to use.

Also check the cost of the surrounding infrastructure when no jobs are running. Scaling the VMs to zero doesn't by itself establish that every supporting component is free.