.png)
This guide explains those cost layers, compares Autopilot with Standard, and shows how GKE pricing differs from Amazon EKS and Azure Kubernetes Service (AKS).
The Short Answer
GKE is a managed Kubernetes service with a $0.10-per-cluster-hour management fee plus charges for the resources running your workloads. Standard normally bills the underlying Compute Engine nodes, while general-purpose Autopilot workloads use Pod-based billing based on requested CPU, memory, and ephemeral storage.GKE is not completely free, although each billing account receives $74.40 in monthly credits for the management fee of Autopilot and zonal Standard clusters. Compute, storage, networking, and regional Standard cluster management fees remain billable.
What Is Google Kubernetes Engine (GKE)?
Google Kubernetes Engine is Google Cloud’s managed service for deploying and operating containerized applications with Kubernetes. Google manages the control plane, while the responsibility for worker infrastructure depends on the mode you use.In Standard mode, your team configures node pools, machine types, scaling, and many workload-level controls. In Autopilot mode, GKE manages nodes and provisions capacity from workload specifications.
Both modes support common uses such as web applications, APIs, microservices, batch processing, and machine-learning workloads, but their operational flexibility and billing behavior differ.
GKE Autopilot vs Standard
The central GKE Autopilot vs Standard decision is whether you want direct control of node infrastructure or want GKE to manage that infrastructure from workload requirements.Google recommends Autopilot as the default mode for most workloads, while Standard remains appropriate when you need configuration or control that Autopilot does not provide. Google’s mode comparison should be checked against your technical requirements before deployment.
| Dimension | GKE Standard | GKE Autopilot |
|---|---|---|
| Infrastructure | You configure and manage node pools | GKE provisions and manages nodes |
| Typical billing basis | Underlying Compute Engine instances | Pod requests for general-purpose workloads |
| Capacity risk | You pay for provisioned nodes, including idle capacity | No charge for unallocated space under Pod-based billing |
| Configuration control | Broad node, machine, and scheduling control | Guardrails and managed defaults |
| Best fit | Specialized or tightly controlled infrastructure | Workloads that fit Autopilot constraints and benefit from lower operational overhead |
How Standard billing works
Standard node pools are billed through Compute Engine until their nodes are deleted. That gives you access to machine-family selection, custom node pools, resource-based CUDs, and detailed scheduling controls, but unused node capacity still costs money.Standard is usually the clearer choice when workloads require specific node configurations, specialized scheduling, or infrastructure-level access. It can also work well for stable workloads when nodes remain highly utilized and commitments match the long-term baseline.
How Autopilot billing works
For general-purpose Autopilot workloads, GKE charges in one-second increments for the CPU, memory, and ephemeral storage requested by running or creating Pods, with no minimum duration. GKE may set defaults or raise requests that do not meet applicable minimums or CPU-to-memory ratios, so accurate resource requests remain essential.Autopilot is not exclusively per-Pod billed. Workloads that select specific hardware, such as a machine series or accelerator, use node-based billing for the provisioned Compute Engine resources plus an Autopilot management premium. See the current GKE Autopilot pricing rules before comparing modes.
What Does GKE Cost?
A GKE bill can contain several independent layers:Cluster management: $0.10 per cluster per hour, billed in one-second increments. Clusters in the Extended release channel can incur an additional fee after entering extended support.
Workload compute: Compute Engine nodes in Standard or the applicable Pod- or node-based Autopilot charges.
Supporting resources: Persistent storage, load balancing, public IP addresses, logging, and network transfer where applicable.
Example monthly calculation
Assume 50 continuously running general-purpose Autopilot Pods in us-central1, each requesting 0.5 vCPU and 1 GiB of memory. Using the published rates of $0.0445 per vCPU-hour and $0.0049225 per GiB-hour:- CPU: 50 x 0.5 x $0.0445 x 730 = $812.13
- Memory: 50 x 1 x $0.0049225 x 730 = $179.67
- Cluster management: $0.10 x 730 = $73.00
- Total before the free-tier credit, storage, and networking: $1,064.80 per month
A Standard comparison requires selecting actual nodes and allowing for system workloads, scheduling headroom, and autoscaling. Use current Compute Engine VM pricing and the Google Cloud Pricing Calculator rather than assuming that the same requested Pod resources translate directly into the same node count.
GKE Discounts: CUDs and Spot Capacity
Commitment eligibility at a glance: Eligible GKE Standard node compute can receive Compute Engine resource-based or Compute Flexible CUDs. For GKE Autopilot, Compute Flexible CUDs provide a 28% discount for a one-year term or 46% for a three-year term on eligible usage.They do not cover cluster management fees, Spot Pods, the Scale-Out compute class, or GPU Pods that do not use the Accelerator class. Because CUDs cannot be cancelled, size the commitment against historical usage and expected minimum eligible hourly spend, not average or peak usage.
For GKE Standard nodes, resource-based CUDs are tied more specifically to resources such as a machine series and region, while Compute Flexible CUDs cover eligible spend more broadly. The right choice depends on how stable the node configuration is. For the full comparison, see Resource-Based vs. Flex CUDs.
For interruptible workloads, Standard supports node pools backed by Spot VMs, while Autopilot supports Spot Pods. Spot prices are dynamic and discounted, but capacity can be reclaimed, so use Spot for fault-tolerant workloads that can stop, retry, or reschedule. Our GKE cost optimization guide covers Spot, rightsizing, and commitment strategy in more depth.
Hidden GKE Cost Drivers
The visible compute rate does not show whether the cluster is using resources efficiently. Review these recurring cost drivers:Idle Standard nodes: Autoscaling cannot eliminate every scheduling gap, system reservation, or capacity buffer. Poor Pod packing can leave paid VM resources unused.
Oversized Autopilot requests: Under Pod-based billing, requested resources influence cost even when actual use is lower. Defaults and minimum adjustments can also affect the bill.
Persistent storage: Volumes can continue generating charges after the workloads that created them are removed, depending on the reclaim policy and deletion workflow.
Network traffic: Cross-zone, cross-region, and internet data transfer can add charges that do not appear in a simple CPU-and-memory estimate.
Too many clusters: Every cluster accrues the management fee; the billing-account credit only offsets eligible Autopilot and zonal Standard cluster fees up to its monthly value.
GKE vs EKS vs AKS Costs
GKE, Amazon EKS, and AKS all separate Kubernetes management from at least some underlying infrastructure charges. Their pricing structures are not identical, so a control-plane fee alone is not a reliable total-cost comparison.| Cost dimension | GKE | Amazon EKS | AKS |
|---|---|---|---|
| Cluster management | $0.10/hour, with an eligible $74.40 monthly credit | $0.10/hour under standard version support; higher during extended support | Free, Standard, or Premium tier; Premium requires AKS Long Term Support |
| Worker compute | Compute Engine nodes or applicable Autopilot billing | EC2, Fargate, or EKS Auto Mode-related charges | Azure VMs and the selected AKS configuration |
| Commitments | Resource-based or Flexible CUDs | Reserved Instances and Savings Plans | Reservations and Azure savings plans for eligible compute |
| Interruptible capacity | Spot VMs and Spot Pods | EC2 Spot | Azure Spot VMs |
For a fair GKE vs EKS vs AKS cost comparison, model the same CPU, memory, availability, storage, traffic, support, and discount assumptions on all three providers. The lowest-cost option often depends more on workload fit, existing cloud commitments, and operating expertise than on the published cluster fee.
How Usage.ai Supports GKE Savings
Long-term discounts can reduce eligible GKE spend, but native CUDs still require a one- or three-year commitment. For teams with a measurable baseline but uncertain future demand, that creates a tradeoff between coverage and underutilization risk.With Flex Insured Commitments, teams can get the 46% discount available with a three-year Compute Flexible CUD on eligible GKE usage with none of the commitment risk.
If a covered Flex Commitment costs more than the equivalent on-demand usage, we calculate that loss and provide cashback protection according to the applicable program terms; accrued cashback is paid after 90 days.
We also charge a percentage of realized savings, aligning our fee with the savings generated through the program.
Frequently asked questions
Is GKE free to use?
Not completely. The monthly credit can offset eligible cluster management fees, but compute, storage, networking, and other services remain billable; regional Standard cluster management fees are excluded from the credit.
What is GKE Autopilot per-Pod billing?
For general-purpose workloads, Autopilot bills the CPU, memory, and ephemeral storage requested by running or creating Pods. Workloads selecting specific hardware use node-based pricing instead, so per-Pod billing is not universal across Autopilot.
How do GKE Spot nodes work?
In Standard, you create node pools backed by Spot VMs; in Autopilot, workloads can request Spot Pods. Both suit fault-tolerant workloads because Google Cloud can reclaim the underlying capacity.
Can Standard workloads use Autopilot features?
Eligible Standard clusters can run workloads that request Autopilot-enabled compute classes, allowing Pod-based Autopilot operation for selected workloads. This is different from converting an entire existing cluster from Standard mode to Autopilot mode.
What is the cheapest small GKE setup?
There is no universal answer. A zonal Standard cluster may benefit from the management-fee credit, while Autopilot can avoid idle-node cost; compare the required availability, Pod requests, node utilization, storage, and traffic before choosing.
