CAST AI

CAST AI replaces the Cluster Autoscaler with an ML provisioner that takes actions, not recommendations.

CAST AI agents watch workload patterns and actively bin-pack pods, rightsize node types, and migrate workloads to spot capacity with sub-minute on-demand fallback when interruption signals arrive. Live migration moves stateful pods between nodes without downtime so the cluster can rebalance continuously. The autonomy level is the differentiator: it is not advising, it is patching the cluster.

Who it's for. Teams running Kubernetes on AWS, GCP, or Azure with at least $10K per month of compute spend. Picture a 50-node EKS cluster mixing Postgres, Kafka, and stateless services: CAST AI live-migrates the stateful pods onto spot-backed nodes, cuts compute by around 60 percent, and falls back to on-demand within 90 seconds when spot capacity tightens.

Tradeoffs. Granting CAST AI write access means it will delete and recreate nodes on its own schedule. That requires real trust and staging validation. Kubernetes only, so ECS and bare VMs are out. The free tier monitors but does not act, and Growth tier pricing is not public.

Compare: Karpenter, StormForge, Spot by NetApp, Zesty