Loading...
Quick Definition
A node is a single machine, physical or virtual, in a cluster that provides compute capacity to run workloads. In Kubernetes, nodes host pods, and node type and count are the main drivers of cluster cost.
A node is a single machine, almost always a virtual machine, that does the work in a Kubernetes cluster. The cluster's brain schedules pods onto nodes; the nodes run them. Everything else in Kubernetes is abstraction, but nodes are physical enough to appear on your invoice.
That is the key cost fact: cloud providers bill nodes by the hour, regardless of how full they are. A node running at 25 percent utilization costs the same as one at 80 percent. Cluster efficiency is therefore largely the art of needing fewer nodes: accurate pod resource requests, tight bin packing, and autoscalers like Karpenter that remove empty machines promptly.
Example. A 40-node cluster averages 30 percent CPU utilization. After requests are rightsized and consolidation is enabled, the same workloads fit on 24 nodes. The bill drops 40 percent and nothing else changes.
Node type choice, size, family, and on-demand versus spot, compounds the effect. Start with node vs pod vs cluster and go deep with the Kubernetes Cost Management Guide.
From a handful to over a hundred, depending on node size and pod resource requests. The scheduler packs by requested capacity.
The share of a node's CPU and memory actually used. Low utilization across a cluster means you are paying for empty space.
Fewer big nodes pack more efficiently with less overhead; more small nodes limit blast radius. Many clusters mix sizes via node pools.