How do I calculate the cost of a Kubernetes pod?

Multiply the pod's vCPU request by the per-core price and the memory request by the per-GB price, then multiply by the number of replicas and by 730 hours per month. A pod requesting 0.5 vCPU and 1 GB of memory at industry-average rates ($21/core/month, $7/GB/month) costs (0.5 x 21 + 1 x 7) = $17.50 per replica per month, or $52.50/month for three replicas.

How much does Kubernetes cost per vCPU and per GB of memory in 2026?

Industry averages across EKS, GKE, and AKS are about $21 per vCPU core per month and $7 per GB of memory per month. By provider: EKS runs about $22/core and $6.50/GB, GKE $21/core and $7/GB, AKS $20/core and $7.50/GB. Managed control planes add roughly $73/month per cluster on EKS and GKE (Azure AKS does not charge for the control plane).

Why does my Kubernetes bill look higher than the pod cost I calculated?

Three reasons: utilization waste (most pods use 20-40% of requested resources, so you pay for requests, not usage), unallocated node capacity (nodes are billed whole even when pods do not fill them), and hidden fees (control plane, load balancers, storage, and egress). A workload that requests $1,000/month at 100% utilization typically delivers only about $300 of real work at 30% utilization.

What is the best way to reduce Kubernetes costs?

Right-size requests first: run kubectl top pods and compare actual usage to requests, then lower over-provisioned limits. Use spot node pools for fault-tolerant workloads (70-80% discounts on AWS EKS, GKE, and AKS). Enable cluster autoscaling so unallocated node capacity shrinks off-peak, and add pod disruption budgets before moving workloads to spot.

How do I convert milliCPU to vCPU for cost calculation?

1,000 millicores equals 1 vCPU. A pod requesting 250m CPU is 0.25 vCPU; 2,000m is 2 vCPU. Convert milli to cores before multiplying by the per-core price: 250m x $21/core/month = $5.25/month per replica. See the milliCPU-to-cores converter for the full table.

Kubernetes Pod Cost Calculator

Pod Spec & Fleet

vCPU request (cores)

Memory request (GiB)

Replicas

Provider pricing

Rates are monthly per core and per GB, based on 2026 industry averages across the big three providers. Convert milliCPU with the milliCPU → cores converter.

Utilization & Overheads

Average utilization

Control plane

📊 Cost Report

Per Pod / Month

Fleet / Month

Fleet / Year

Effective Work

Waste / Month

Utilization

Hourly (fleet)

📋 Managed Kubernetes Pricing Reference (2026, monthly)

Provider	CPU / core	Memory / GB	Control Plane
AWS EKS	$22	$6.50	$73/cluster
Google GKE	$21	$7	$73/cluster
Azure AKS	$20	$7.50	Free
Industry avg	$21	$7	~$73

The Pod Cost Model: Requests, Not Usage

Kubernetes bills you for what pods request, not what they use. The node is paid whole, and the pod cost is its share of that node — which is why the two numbers that matter are the vCPU and memory requests in the pod spec. The formula is the same one OpenCost, Kubecost, and Harness use: (vCPU × per-core price) + (memory GB × per-GB price), scaled by replicas and the 730 hours in a month. A microservice requesting 0.5 vCPU and 2 GiB across three replicas lands at roughly $52.50/month before anything else.

The Waste Factor

Most pods use 20–40% of what they request. A fleet that requests $1,000/month of resources and runs at 30% utilization is doing about $300 of real work — the other $700 is idle allocation you are paying for. The fix is not more hardware; it is right-sizing: run kubectl top pods, compare actual usage to requests, and cut the over-provisioned limits. This is the single largest Kubernetes saving available, and it costs nothing but an afternoon. If you are also wondering how many pods actually fit on a node, the Pod Density briefing covers why the default 110-pod cap is a lie at production workloads.

Spot, Control Planes, and the Hidden Line

Fault-tolerant workloads (batch, CI/CD, stateless APIs with replicas) can run on spot node pools at 70–80% discounts — add a pod disruption budget before you do. And remember the control plane: EKS and GKE charge roughly $73/month per cluster even if it runs one pod, so consolidating small clusters is often free money. For the wider Kubernetes cost picture, the container resource limit calculator and the Cloud Quotas toolbox cover adjacent sizing decisions.

Pricing Basis, Sources & Assumptions

Every rate on this page is a published vendor list price — no negotiated discounts, private pricing, or credit offsets. Totals are in USD and exclude tax. Rates were last checked against the sources below on 2026-09-23.

Pricing input	Basis used on this page
Region	Blended rates for AWS, Google Cloud, and Microsoft Azure in their main US regions (us-east-1, us-central1, East US), selected by the provider preset.
Currency	USD — on-demand list price, tax excluded
Last checked	2026-09-23 — The provider preset rates are blended per-core and per-GB figures derived from published node pricing as of this date — they are not a single provider SKU price, so re-check node pricing before committing to a specific instance family.

Pricing input

Basis used on this page

Region

Blended rates for AWS, Google Cloud, and Microsoft Azure in their main US regions (us-east-1, us-central1, East US), selected by the provider preset.

Currency

USD — on-demand list price, tax excluded

Last checked

2026-09-23 — The provider preset rates are blended per-core and per-GB figures derived from published node pricing as of this date — they are not a single provider SKU price, so re-check node pricing before committing to a specific instance family.

Modelling assumptions

Rates are per vCPU and per GiB of memory per month, so a pod is charged for what it requests, not for what it uses.

Provider presets: industry average $21 per core + $7 per GB; AWS EKS $22 + $6.50; Google GKE $21 + $7; Azure AKS $20 + $7.50.

Memory is binary throughout, matching Kubernetes resource units: 1 GiB = 1,024 MiB.

The control-plane charge is added separately when that toggle is on — managed control planes bill per cluster-hour, not per node.

Utilisation only drives the waste line. It does not reduce the bill, because Kubernetes reserves the full request whether the container uses it or not.

A full month is 730 hours.

What this model excludes

Spot, preemptible, Savings Plan, and committed-use rates, which can cut the node rate by 30–70%.

Idle capacity created by the cluster autoscaler, and Fargate-style per-pod pricing.

Persistent volumes (EBS/PD/disk), data transfer, and observability agents that run one per node.

Reserved system capacity, which makes allocatable lower than the node's nominal size.

Sources

AWS (2026). "Amazon EC2 Instance Types." Instance families and their vCPU-to-memory ratios behind the AWS preset. aws.amazon.com

Google Cloud (2026). "GKE Cluster Architecture." Node and control-plane architecture behind the Google Cloud preset. cloud.google.com

Microsoft Azure (2026). "AKS Concepts — Networking." Node-pool architecture behind the Azure preset. learn.microsoft.com

Kubernetes (2026). "Managing Resources for Containers." Requests versus limits, and the resource units (Ki/Mi/Gi) used on this page. kubernetes.io

Kubernetes (2026). "Reserve Compute Resources for System Daemons." Why allocatable capacity is always lower than node capacity. kubernetes.io

Vendor list prices change without notice — re-check the linked pages before committing spend. jslet takes no vendor sponsorship and carries no affiliate links; see about.