Kubernetes 1.36 reached general availability upstream in April 2026. By late June, the three largest managed Kubernetes services had each brought it to customers, but not on the same terms. Amazon shipped full support first. Microsoft designated it a long-term support release. Google made it available only in its pre-SLA preview channel. The gap between "supported everywhere" and "available if you accept the risk" says more about each vendor's AI infrastructure priorities than any single feature in the release.
AWS moves first, across every region
Amazon Web Services announced on June 2, 2026 that Amazon EKS and Amazon EKS Distro now support Kubernetes 1.36, available in every AWS region where EKS runs, including the GovCloud (US) regions (AWS, 2026). Customers can create new 1.36 clusters or upgrade existing ones through the EKS console, eksctl, or infrastructure-as-code tooling immediately.
The release itself carries features aimed squarely at AI and mixed-workload clusters. User Namespaces reached general availability, letting a container's root user map to an unprivileged account on the host node, a meaningful isolation improvement for multi-tenant clusters running untrusted or third-party workloads. Mutating Admission Policies became available for CEL-based resource mutation directly in the API server, removing the need to run separate webhook infrastructure just to enforce or rewrite pod specs. In-place pod-level vertical scaling lets a running pod resize its CPU and memory budget without a restart, useful for long-running training or inference jobs that would otherwise lose state on a resize. And a new Resource Health Status field reports device-level health in pod status, giving cluster operators visibility into failing GPUs or other accelerators without building custom tooling (AWS, 2026).
AWS lists standard support for 1.36 running through April 2027, with extended support to June 2027, following its usual 14-month standard window from upstream GA.
Azure designates it long-term support
Microsoft's rollout followed a slower, staged path: an upstream release in April, an AKS preview in May, and general availability on June 19, 2026, when 1.36 was designated a Long Term Support version (Azure/AKS, 2026; Microsoft Learn, 2026). Clusters no longer need an opt-in preview flag to create or upgrade to 1.36, a meaningful signal that Microsoft considers the release stable for production.
AKS layered its own additions on top of the upstream release. FIPS-compliant node pools became available on Ubuntu 22.04, relevant to regulated customers with federal or financial compliance requirements. New VM sizes built around NVIDIA's RTX PRO 6000 GPU shipped alongside it. Deployment Safeguards, AKS's policy-enforcement layer, extended its default resource-request behavior to DaemonSets and Jobs, not just Deployments and StatefulSets, closing a gap that had let those workload types skip resource limits entirely. AKS Automatic clusters also gained the option to run Istio-based service mesh with Istio CNI in place of the default application routing add-on (Azure/AKS, 2026).
The LTS label matters more than any individual feature. It tells platform teams they can plan a multi-year upgrade cadence around 1.36 rather than treating it as a release to skip past quickly.
Google holds 1.36 back from its supported channels
Google's GKE documentation shows a different posture entirely. As of late June, 1.36 was available only in the Rapid channel, GKE's fastest and least stable release track. Google's own release notes state plainly that Rapid channel versions are "excluded from the GKE SLA and may contain issues without known workarounds" (Google Cloud, 2026). The Regular and Extended channels, the ones most production clusters run on, were still defaulting to 1.35 patch versions during this window.
That is not a delay by accident. GKE's standard practice is to prove a new minor version in Rapid before promoting it, and Google was in the middle of that process rather than skipping it. But set against Amazon's day-one, all-region support and Microsoft's LTS designation, it puts Google noticeably behind on this particular release, a reversal of the position Google typically holds as the vendor with the deepest in-house Kubernetes expertise.
The AI conformance program keeps expanding underneath all three
A separate but related thread: the Cloud Native Computing Foundation's Certified Kubernetes AI Conformance Program, launched in November 2025 to define what "AI-ready" actually means across competing Kubernetes distributions (CNCF, 2025), nearly doubled its certified platform count in the months since, from 18 platforms to 31 as of a March 2026 update (CNCF, 2026). The program exists because "runs Kubernetes" had stopped being a useful signal for buyers evaluating whether a platform could actually schedule GPUs, report accelerator health, or isolate multi-tenant AI workloads safely. The features landing in 1.36, resource health status and in-place scaling among them, are the raw material that conformance testing checks for.
What this means for buyers, not just operators
None of this is abstract for anyone running production AI workloads on managed Kubernetes. A team evaluating GKE for a GPU-heavy training pipeline in June 2026 was choosing between staying on 1.35 in a supported channel or moving to 1.36 in an unsupported one, months before AWS or Azure customers faced that tradeoff. A team on AKS got an LTS commitment on a release with meaningfully better GPU health visibility. The three hyperscalers are converging on the same set of AI-native primitives, but the pace and the risk each one asks customers to absorb to get there still differ sharply.
What to watch next
GKE's Regular channel promotion of 1.36 is the next concrete milestone. When Google moves it out of Rapid and into a supported channel with an SLA attached, it will close the gap that opened in June. Watch also for the CNCF's next conformance count update, expected in the second half of 2026, as an indicator of how quickly the smaller Kubernetes distributions are catching up to the big three on AI-specific capability.
Sources: AWS · Azure/AKS · Microsoft Learn · Google Cloud · CNCF launch · CNCF update