Kubernetes as the operating model

Kubernetes started as a traffic controller for containers: you tell it how many copies of an app you want, and it keeps that number running across a pool of machines. The official project still defines it as software for deploying, scaling, and managing containerized applications. (kubernetes.io) The key trick is the control plane, which is the part that watches the real world and keeps pushing it toward the state you declared in code. If one container crashes, Kubernetes notices the gap and starts another one without waiting for a human. (kubernetes.io) For years, that control plane mostly stayed inside one cluster, which is one fenced-off group of machines managed as a unit. The new shift is that teams are using the same control-plane pattern to manage many clusters and even resources outside clusters. (cluster-api.sigs.k8s.io, karmada.io) You can see that in Cluster Application Programming Interface, an official Kubernetes subproject that uses Kubernetes-style application programming interfaces to create, upgrade, and operate multiple clusters. It turns cluster management itself into something you declare in files instead of click through in consoles. (cluster-api.sigs.k8s.io, kubernetes.io) You can see it again in Karmada, a Cloud Native Computing Foundation project built for multi-cloud and multi-cluster orchestration. Karmada says it can run applications across multiple Kubernetes clusters and clouds with centralized management, failure recovery, and traffic scheduling. (karmada.io) And you can see the boundary moving in the other direction too, into virtual machines, which are software-made computers that behave like separate servers. KubeVirt adds new Kubernetes application programming interface types so operators can run and manage virtual machines through the same Kubernetes control surface they use for containers. (kubevirt.io, kubevirt.io) Crossplane pushes the idea further by treating Kubernetes as a way to manage external infrastructure, not just workloads inside a cluster. Its documentation describes a control plane that can orchestrate applications and infrastructure no matter where they run. (docs.crossplane.io, github.com) That is why the newest vendor news is not really about another cluster feature. At KubeCon + CloudNativeCon Europe in Amsterdam on March 23 to March 26, 2026, anynines showed Klutch and a9s Hub as a Kubernetes-based control plane for data services across on-premises systems, Amazon Web Services, Cloud Foundry, and virtual-machine environments. (events.linuxfoundation.org, efficientlyconnected.com) The important detail is the target: data services like databases, storage, and messaging have usually been managed in separate tools from the application platform. Efficiently Connected described that layer as one of the most fragmented parts of hybrid and multi-cloud operations, which is why companies are trying to put it behind one Kubernetes-native interface. (efficientlyconnected.com) KubeCon’s own framing shows how much the conversation has changed. A Cloud Native Computing Foundation post on April 10 said Platform Engineering Day had reached its fifth edition and focused on internal platforms, production operations, and guardrails for artificial intelligence and autonomous agents, not just cluster setup. (cncf.io) Amazon Web Services said the same thing more bluntly before the conference: platform engineering has become the operating model for scaling Kubernetes across diverse workloads, multiple clusters, and multiple environments. In the 2025 Cloud Native Computing Foundation survey released on January 20, 2026, 82 percent of container users said they now run Kubernetes in production. (aws.amazon.com, cncf.io) So the story is no longer “Kubernetes won containers.” The story is that Kubernetes is being used like a common operating model for messy estates that mix containers, virtual machines, bare metal, clouds, and stateful data services, while platform teams decide who gets self-service access and which policies every request has to pass. (aws.amazon.com, crossplane.io, efficientlyconnected.com)