Why this release matters

Kubernetes is already the default for container orchestration, but the goal is no longer only scale. The goal is specialization.

v1.36 reinforces the cluster’s ability to understand workload profile and behavior more precisely.

That directly affects teams building ML platforms, multi-tenant environments, and large-scale operations.

How Kubernetes is evolving

Over recent cycles, Kubernetes has moved from a generic platform toward infrastructure that is more intelligent and more opinionated where it matters.

  • Better support for specialized hardware (GPUs, accelerators)
  • Stronger isolation and security controls
  • Less dependence on external glue

v1.36 consolidates that direction.

70 enhancements and stage changes

In Kubernetes, enhancements are tracked as proposals (KEPs) and move through Alpha, Beta, and Stable/GA stages. They include new APIs, feature gates, security changes, and evolutions in scheduler, kubelet, storage, and metrics.

v1.36 Haru ships 70 enhancements: 18 Stable, 25 Beta, and 25 Alpha. For the full technical list, refer to CHANGELOG-1.36.

Stable (GA) highlights

Promotions to GA in this cycle include:

  1. Support User Namespaces in pods
  2. API for external signing of Service Account tokens
  3. Speed up recursive SELinux label change
  4. Portworx file in-tree to CSI driver migration
  5. Fine grained Kubelet API authorization
  6. Mutating Admission Policies
  7. Node log query
  8. VolumeGroupSnapshot
  9. Mutable CSINode Allocatable Property
  10. DRA: Prioritized Alternatives in Device Requests
  11. Support PSI based on cgroupv2
  12. add ProcMount option
  13. DRA: Extend PodResources to include resources from Dynamic Resource Allocation
  14. VolumeSource: OCI Artifact and/or Image
  15. Split L3 Cache Topology Awareness in CPU Manager
  16. DRA: AdminAccess for ResourceClaims and ResourceClaimTemplates
  17. Remove gogo protobuf dependency for Kubernetes API types
  18. CSI driver opt-in for service account tokens via secrets field

Beta — widely tested

  • Pod allocated device health status (KEP 4680)
  • Controller staleness mitigation (KEP 5647)
  • Strict IP/CIDR validation (KEP 4858)
  • kubectl user preferences in .kuberc (KEP 3104)
  • Restricted impersonation (KEP 5284)
  • DRA improvements across multiple KEPs (5004, 4817, 5055, 5075, 4815, and 5007)

Alpha — experimental

  • HPA scale-to-zero with custom metrics (KEP 2021)
  • Workload Aware Scheduling in evolution (KEPs 4671, 5547, 5832)
  • DRA extensions for additional scenarios (for example 5729, 5304)
  • Native histogram support for metrics (KEP 5808)

Main changes in v1.36

In practice, these themes concentrate most of the impact for the majority of clusters:

  • Workload Aware Scheduling (WAS) improves scheduling decisions.
  • Dynamic Resource Allocation (DRA) advances support for GPU and AI workloads.
  • Stronger security with user namespaces and more granular policy controls.
  • Removal/deprecation of legacy features such as gitRepo and externalIPs.

Workload Aware Scheduling in practice

Traditionally, Kubernetes schedules pods individually. WAS enables multiple pods to be treated as a single logical unit.

That improves:

  • consistency for distributed applications
  • execution predictability
  • resource efficiency

For complex workloads (such as data pipelines and ML), this is a significant shift.

DRA and AI-style workloads

DRA keeps evolving as the foundation for specialized resources.

As a result, Kubernetes is better positioned to support:

  • shared GPU usage
  • accelerator-specific allocation
  • demand-based dynamic allocation

This is one reason Kubernetes is becoming a practical default for AI workloads in production.

Operational impact

v1.36 reduces dependence on external tooling and increases native platform capability.

  • less dependence on custom operators
  • better native observability
  • tighter isolation controls

For platform engineering teams, that means more control with lower operational complexity.

Conclusion

Kubernetes v1.36 is not just an incremental update. It is a clear step toward a platform that is smarter, safer, and better prepared for modern workloads.

This release is especially relevant for:

  • platform teams
  • cloud-native engineers
  • teams building and operating AI/ML systems

If you want the short story first, read the news summary.