Kubernetes v1.35 ('Timbernetes') introduces several features that strengthen its role as the operational substrate for AI/ML workloads. Key changes include workload-aware scheduling with gang scheduling support (alpha) for distributed training, stable in-place Pod resource resize to reduce restart churn in inference services, continued Dynamic Resource Allocation (DRA) progress for GPU/accelerator orchestration, and KYAML as the new default kubectl output format for safer configuration workflows. The post also flags the Ingress NGINX retirement timeline (March 2026) as a planning item for operators, and frames these updates as Kubernetes absorbing more AI infrastructure complexity over time.
Table of contents
Why v1.35 reads like an AI-infrastructure releaseThe changes that matter for AI/ML operationsIn-place Pod resize is stableDevice allocation keeps moving toward a baseline capabilityKYAML becomes the default kubectl output formatWhy AI keeps pushing teams toward a shared operating layerPlatform engineering implicationsEcosystem note: Ingress NGINX retirement timelinePractical evaluation steps for v1.35About the author18.1K Impressions