Ch 5 · GKE & Serverless
Billing model is the decision
23 / 44

Autopilot bills per Pod; Standard bills per node, always.

Autopilot now mandates resource requests on every Pod.

StandardAutopilot
BillingPer node VM, 24/7Per Pod request, only while running
Pod requestsOptional (not advised)Mandatory, defaults injected
Best fitHigh, consistent utilizationLow or bursty utilization
Migrating a bursty reporting service to Autopilot on a general "serverless is cheaper" heuristic backfired — the bill came in 40% higher, because the workload ran at consistently high CPU whenever active. Check measured utilization, not a rule of thumb.