Datadog Launches Fleet-Wide Kubernetes Autoscaling at Scale
Datadog has introduced new fleet-wide deployment options for its Kubernetes Autoscaling product, giving platform teams a faster and more centralized way to rightsize workloads across their entire infrastructure. The update focuses on three rollout paths: in-app setup, GitOps cluster profiles, and AI-assisted pull request generation. Together, they make Kubernetes autoscaling at scale accessible without burdening individual application teams.
Every Kubernetes environment accumulates waste over time. Teams overprovision CPU and memory to avoid performance risk. They also run idle replicas long after workload behavior has changed. While Datadog Cluster Autoscaling already addresses node-level waste, the biggest savings often hide at the workload level, where resource requests and replica counts are set service by service.
The new in-app setup workflow addresses this directly. From the autoscaling setup page, platform teams can view workloads across a cluster, see which are ready for immediate activation, and deploy autoscalers in bulk, all without writing YAML. The page also surfaces estimated idle costs and potential savings, so teams can prioritize high-impact workloads first.
For teams that manage infrastructure through Git, cluster profiles offer a cleaner path. A single DatadogPodAutoscalerClusterProfile custom resource can define autoscaling policy once, then apply it across namespaces with a label. The Datadog Cluster Agent detects the label and automatically creates the right autoscaler resources. As a result, adding autoscaling to a namespace can be as simple as a one-line pull request. Because everything lives in Git, the change follows the same review and approval process as any other infrastructure update.
The third path uses AI to speed up onboarding. Teams can run AI-assisted setup from the Datadog UI via Bits AI Dev or from tools like Claude, Cursor, or Codex through the Datadog Model Context Protocol (MCP) Server. The assistant inspects the cluster, reviews any existing HPAs, WPAs, or VPAs, and then generates equivalent DatadogPodAutoscaler manifests. It even prepares a draft PR for review. This is especially useful for teams making their first autoscaling change, since the hardest part of any rollout is often simply opening that first PR.
Across all three rollout paths, the Datadog Pod Autoscaler also supports in-place vertical resizing. Rather than requiring pod recreation to adjust CPU or memory requests, supported changes are applied directly to running containers. This reduces disruption and removes one more reason teams delay autoscaling adoption. Stale resource requests no longer need a full redeploy cycle to fix.
Kubernetes autoscaling at scale becomes practical when platform teams can manage policy centrally rather than negotiating configuration with every service owner. Datadog’s three-path approach, in-app, GitOps, and AI-assisted, gives teams the flexibility to match their existing deployment workflow while still reducing idle costs fleet-wide.
Full documentation for Kubernetes Autoscaling and idle cost and savings estimates is available on the Datadog docs site.
