Cloudchipr splits each node's amortized hourly cost across the pods it runs, in proportion to requested CPU and memory, then rolls that up by workload, namespace, node, and cluster.

How it works, step by step

1. Start from each node's amortized cost

Every node has an amortized hourly cost, meaning the effective price you pay once Spot discounts and commitments are spread across usage. A Spot node is costed at its Spot rate; an On-Demand node at its On-Demand rate. No list-price estimates are used.

2. Split the node price into CPU and memory

A node's hourly cost is divided between the CPU it provides and the memory it provides. This gives every CPU-hour and every GB-hour on that node a price.

3. Allocate to workloads by what they request

Each pod, workload, and namespace is charged for the CPU and memory it requests, as its share of the node it runs on. Request more, pay more; request less, pay less.

4. Add it up over time

Costs are summed across every hour your workloads ran, then rolled up by workload, namespace, node, and cluster, including nodes that scaled up for an hour and disappeared.

Where a node's cost goes

One node's hourly cost equals 100% of the price you pay, and it is divided into two kinds of line items:

The workload slices plus the idle slice always add up to exactly what the node costs. Nothing is invented and nothing is lost, which is why idle cost is a reliable signal for right-sizing.

Why the price is often lower than an On-Demand estimate

Amortized cost includes your Spot savings. Clusters that use Spot capacity are billed at the Spot rate, which is typically 50 to 70% below the On-Demand list price for the same hardware. If you estimate cost from On-Demand rates, Cloudchipr will look lower, because you're only paying the Spot price.

Example node: 16 vCPU / 64 GB

How you see it in Billing Explorer

Kubernetes cost appears alongside the rest of your cloud spend, and you can break it down the way your teams are organized.

Group and filter. Break cost down by cluster, namespace, workload, node, or label, and filter to any team, environment, or time range.

CPU and memory, separately. Each workload shows its CPU cost and its memory cost as distinct components, so you can see what's driving the bill.

Idle cost in the open. Idle node capacity is its own line rather than being spread across workloads, which makes over-provisioning easy to spot and reduce.

Shared clusters, fair splits. On a shared cluster, each namespace pays only for the resources it requests. One team's usage never inflates another's cost.