GKE Autopilot
A mode of Google Kubernetes Engine where Google manages node provisioning, sizing, and patching entirely - you define workloads, and nodes appear automatically to run them, billed per pod resource request rather than per node. It is the lower-effort default for most containerized workloads, versus Standard mode's direct node pool control.
Frequently Asked Questions
How does GKE Autopilot's billing model differ from Standard mode?
Standard mode bills for the underlying VM nodes you provision, whether or not pods fully utilize them — idle node capacity is still paid for. Autopilot instead bills per pod based on the CPU, memory, and storage the pod actually requests, with Google handling node sizing and bin-packing behind the scenes. This removes node-level capacity planning but means costs track workload requests directly, so over-requesting resources in a pod spec has a more immediate, visible cost impact.
What trade-off should a team weigh before choosing Autopilot over Standard mode?
Autopilot restricts certain configurations Standard mode allows — like DaemonSets with privileged access, custom node-level kernel tuning, or specific third-party CNI/storage plugins — since Google manages the node layer directly. Teams with workloads needing that low-level node control (specialized GPU drivers, host networking, custom sysctls) should stick with Standard; most stateless application workloads without such requirements are a good fit for Autopilot's reduced operational overhead.