Kubernetes is the new platform
A recent 'State of Kubernetes' update argues Kubernetes is no longer just an orchestrator but the operating model around which platform and AI teams organize developer workflows and service delivery. That shift is showing up in operational demands — consistent lifecycle management, GPU scheduling, observability and multi‑cluster APIs — which turns Kubernetes into the integration point between VMs, bare metal and cloud. For teams running VCF and vSphere, the challenge is to make VMware infrastructure the best substrate for those Kubernetes‑centric platform patterns rather than merely supporting containers as an afterthought. (www.youtube.com (x.com)
Kubernetes started as a traffic controller for containers, which are apps packed with everything they need to run. In 2026, platform teams increasingly use it more like a city hall that also handles permits, roads, power, and zoning for the whole software estate. (kubernetes.io) (cluster-api.sigs.k8s.io) A container is a sealed shipping box for software. Kubernetes decides which machine gets each box, restarts it if it crashes, and spreads copies across servers so one hardware failure does not take the app down. (kubernetes.io 1) (kubernetes.io 2) That was enough when companies were mostly trying to stop hand-configuring virtual machines one by one. It is not enough when one team needs a database, another needs an application programming interface gateway, and an artificial intelligence team needs eight graphics processing units on the same cluster by 9 a.m. (kubernetes.io 1) (kubernetes.io 2) Graphics processing units are the chips that train and run many artificial intelligence models, and Kubernetes now has stable support for scheduling them as cluster resources. It does that through device plugins, which are vendor add-ons that tell Kubernetes exactly which special hardware each server has. (kubernetes.io 1) (kubernetes.io 2) (github.com) Once a company starts treating Kubernetes as the place where scarce chips get assigned, it usually wants the same system to handle upgrades too. Cluster API, a Kubernetes subproject, was built for that job: create, scale, and upgrade whole Kubernetes clusters using the same declarative model people already use for apps. (cluster-api.sigs.k8s.io) (github.com) Then comes observability, which is the practice of collecting metrics, logs, and traces so operators can see what is breaking before users do. Kubernetes documents now describe those three signals as the standard way to understand cluster health, which is why platform teams keep piling monitoring and policy systems around the cluster instead of beside it. (kubernetes.io) The next step is multi-cluster management, which means one control layer for many Kubernetes fleets across regions or environments. Broadcom’s VMware Cloud Foundation team said in August 2025 that active VMware Cloud Foundation subscribers would get Tanzu Mission Control Self-Managed for Kubernetes fleet management at no extra cost, starting in May 2025. (blogs.vmware.com) That tells you what changed inside VMware’s world. The pitch is no longer “we can run containers on virtual machines,” but “we can make virtual machines, storage, networking, and graphics processing units look like one clean Kubernetes platform for platform engineers.” (vmware.com) (techdocs.broadcom.com) VMware’s newer vSphere Kubernetes Service materials make that point directly by focusing on conformant Kubernetes, two-click cluster deployment, and upgrade guardrails tied into VMware Cloud Foundation. In other words, the infrastructure is being sold as the concrete foundation under a Kubernetes-shaped building, not as the building itself. (techfieldday.com) (youtube.com) That is why people now say Kubernetes is the platform. The software teams ask for a cluster, a policy, a service endpoint, or a graphics processing unit quota in Kubernetes terms first, and the virtual machines, bare metal servers, and cloud accounts underneath increasingly just exist to satisfy that request. (cluster-api.sigs.k8s.io) (kubernetes.io) (vmware.com)
