Grafana Labs transitioned from using Cluster Autoscaler (CA) to Karpenter for managing its Amazon Elastic Kubernetes Service (EKS) clusters to address inefficiencies and reduce both costs and complexities. Initially, CA was chosen for its simplicity and proven track record, but it soon became evident that it was not optimal for Grafana's needs due to issues like underutilized nodes, limited flexibility with instance types, and difficulties with using AWS Spot Instances. Grafana Labs found Karpenter, an open-source solution primarily developed by AWS, to be more aligned with their requirements because of its just-in-time node provisioning, native support for capacity type prioritization, and its ability to directly interact with AWS EC2 API for optimal instance selection. This switch resulted in a significant reduction in idle capacity and costs, improved reliability, and better developer experience without disrupting service to users. Although Karpenter is still under active development and currently limited to AWS, the benefits it brought to Grafana Labs, such as enhanced disaster recovery capabilities and simplified infrastructure management, outweighed the trade-offs of adopting a cloud-specific scaling solution.