Kubernetes v1.37 'Garhwal' Adds Native Scale-to-Zero for HPA

Kubernetes released version 1.37, codenamed 'Garhwal', on August 26, 2026. The update introduces critical improvements to resource management and cluster reliability designed to reduce operational overhead for platform engineers.

The release consists of 67 total enhancements, comprising 16 Stable (GA), 23 Beta, and 27 Alpha features, alongside one removal or deprecation. A primary highlight is the introduction of native scale-to-zero support for the HorizontalPodAutoscaler (HPA). Now available in Beta and enabled by default, this feature allows workloads to scale down to zero pods when using object or external metrics—a capability that previously often required the deployment of external tools like KEDA.

Maturing Resource Allocation

This release follows the project's established cycle of graduating features from Alpha and Beta stages to Stable. A significant milestone in this version is the stabilization of the core of Dynamic Resource Allocation (DRA), with DRA Extended Resource support officially reaching General Availability. By maturing how the orchestrator handles specialized hardware and dynamic requests, Kubernetes is moving toward a more flexible model for high-performance computing and AI workloads.

Enhancing Cluster Stability

Beyond resource allocation, v1.37 addresses critical stability concerns regarding the API server. The API server has been improved to prevent the overloading of etcd during startup. This change mitigates a known risk where restarting the API server could trigger a cascade of requests that overwhelm the backend data store, potentially leading to cluster-wide instability during recovery phases.

Industry Implications

For infrastructure operators, the native scale-to-zero capability in HPA is a significant cost-saving measure. By eliminating idle replicas during periods of zero demand, organizations can substantially lower their cloud compute spend without sacrificing the ability to react to incoming traffic. Combined with the stabilization of DRA, these updates allow for more precise control over expensive hardware resources, ensuring that GPUs or other accelerators are only consumed when actively needed.

What to Watch

As the community adopts 'Garhwal', the focus will shift toward the 27 Alpha features currently in the pipeline, which will determine the trajectory of the next few releases. Operators should monitor the transition of these experimental features into Beta to plan their long-term infrastructure roadmaps. Additionally, the industry will be watching how the native HPA scale-to-zero implementation performs at scale compared to third-party autoscaling frameworks.