How we saved over $3 million in idle compute costs with Datadog Kubernetes Autoscaling

Datadog's introduction of Datadog Kubernetes Autoscaling (DKA) has significantly optimized its Kubernetes resource management by intelligently automating both horizontal and vertical workload scaling. The implementation has helped eliminate over $3 million in annual idle compute costs, while improving reliability across its services. A central part of this success was demonstrated by Rapid, a platform team at Datadog, which adopted DKA to manage over 1,800 services and 20,000 deployments. DKA's multidimensional scaling mode addressed previous challenges with separate autoscaling tools, simplifying configurations and reducing both overprovisioning and underprovisioning issues. This allowed Rapid to fine-tune resource allocations based on actual usage, cutting costs by over 50% in some data centers and improving service reliability. DKA's successful rollout not only streamlined resource management but also fostered a culture of cost ownership among Datadog's engineering teams, enabling them to optimize services and share improvements company-wide, creating a cycle of continuous enhancement.