Why the platform runs on Karpenter
Choosing a node autoscaler for bursty, GPU-heavy workloads.
- status
- accepted
- decided
- q3 2025
- considered
- cluster-autoscaler · custom controller
- systems
- aws · karpenter · kubernetes
decision
Standard Cluster Autoscaler provisions identical node groups. Karpenter responds directly to unschedulable pod requirements, selecting the cheapest GPU instance family that satisfies the memory specification.