EKS vs AKS vs GKE: Managed Kubernetes Pricing Compared (2026)

AWS EKS and GKE Standard both charge $0.10/hr ($73/mo) for the control plane, while AKS includes it at no extra cost (AWS EKS Pricing, Azure AKS Pricing, GKE Pricing). But the control plane fee is a small fraction of your total spend.

The table reveals an important pattern: list prices for nodes, load balancers, and NAT gateways are remarkably similar across all three providers. The differences that matter come from discount structures, networking topology, and how each provider handles autoscaling. Don't pick a provider based on the control plane fee alone.

As of March 2026, AWS EKS charges $0.10 per hour for the Kubernetes control plane, totaling $73 per month per cluster. Azure AKS provides the control plane at no charge. GKE Standard charges $0.10 per hour, while GKE Autopilot waives the control plane fee but adds a per-pod resource surcharge (Google Cloud, 2026 ).

The control plane runs your API server, etcd, scheduler, and controller manager. EKS and GKE Standard each charge $0.10/hr — roughly $73/month — while AKS bundles it free with every cluster (Azure AKS Pricing). For a single cluster, that's $876/year you save on AKS versus the other two.

That said, AKS's free tier comes with a weaker SLA. The free tier offers a 99.5% uptime target, while the paid "Standard" tier at $0.10/hr matches EKS and GKE at 99.95%. So if you need production-grade availability, the control plane price converges across all three providers. The "AKS is free" headline is only half the story.

EKS Extended Support adds hidden costs

AWS introduced Extended Support for EKS in 2024, charging $0.60/hr per cluster for Kubernetes versions that have left standard support (AWS EKS Pricing). That's $438/month on top of the $73 base fee. If your team is slow to upgrade, EKS becomes significantly more expensive than AKS or GKE. Staying on a supported version isn't optional — it's a cost control decision.

GKE Autopilot changes the model

GKE Autopilot doesn't charge a separate control plane fee. Instead, you pay a premium on pod-level resources — roughly 20-30% more per vCPU and per GB of memory compared to GKE Standard node pricing (GKE Pricing). For clusters with variable workloads and low average utilization, Autopilot can actually cost less because you don't pay for idle node capacity.

Node compute is the largest cost driver, accounting for 60-70% of a typical managed Kubernetes bill according to CAST AI's 2025 Kubernetes Cost Report. Across equivalent instance types, on-demand prices fall within 5-15% of each other. The real savings come from discount programs.

On-demand pricing for a 4 vCPU / 16 GB node

GKE's e2-standard-4 runs about 17% cheaper than EKS's m7i.xlarge on list price. But instance families aren't perfectly equivalent. The m7i uses 4th Gen Intel Xeon processors; the e2 uses a mix of Intel and AMD. For CPU-bound workloads, you may need a c3 or n2 instance on GCP, which narrows the gap.

Discount programs make the real difference

AWS Savings Plans offer up to 72% off on-demand pricing with a 3-year commitment (AWS Savings Plans). Azure Reservations offer similar discounts up to 72% for 3-year terms. GCP stands apart with automatic sustained-use discounts — up to 20% off with no commitment, applied to any instance running more than 25% of the month (Google Cloud Docs). GCP Committed Use Discounts add up to 57% for 3-year terms.

Here's what matters for small teams: GCP's sustained-use discounts are automatic. You don't need to forecast capacity or commit upfront. AWS and Azure require you to actively purchase reservations or Savings Plans. If your team doesn't have a FinOps practice, GCP's default discount behavior quietly saves you 20% that you'd otherwise leave on the table.

Spot and preemptible instances

All three providers offer deeply discounted interruptible instances. AWS Spot Instances save 60-90% off on-demand. Azure Spot VMs save up to 90%. GCP Spot VMs (formerly preemptible) save 60-91% (GCP Spot VMs). For stateless workloads and batch jobs on Kubernetes, spot instances are the single biggest cost lever available.

NAT gateways, load balancers, persistent volumes, and data egress add 30-50% on top of raw node costs for typical clusters. In our experience, these "ancillary" charges surprise teams more than the compute bill itself. A Datadog State of Cloud Costs report found that 83% of container spend is idle, which compounds these hidden costs further.

For guidance on attributing these costs to teams, see our complete guide to Kubernetes cost allocation.

NAT gateway costs

Every private subnet that needs outbound internet access requires a NAT gateway. All three providers charge roughly $32/month as a base fee, plus $0.045/GB of processed data. A cluster pulling container images, calling external APIs, and shipping logs can easily process 500 GB/month — adding $54.50 on top of the base fee. That's $86.50/month per AZ that many teams overlook.

AWS NAT Gateway pricing is particularly painful because you pay both the processing fee and the standard data egress charge (AWS VPC Pricing). A single NAT gateway processing 1 TB/month costs $77.40 in gateway fees alone, before egress. We've seen teams whose NAT costs exceed their EKS control plane fee by 2x.

Load balancer costs

Each Kubernetes Service of type LoadBalancer provisions a cloud load balancer. On AWS, an Application Load Balancer (ALB) costs $16.20/month plus $0.008 per LCU-hour. Azure's Standard Load Balancer runs $18/month. GKE's load balancer costs about $18/month. The catch: many teams create multiple load balancers for different services instead of using an Ingress controller. Three load balancers cost $48-66/month across any provider.

Persistent volume costs

Kubernetes persistent volumes (PVs) map to cloud block storage. AWS EBS gp3 costs $0.08/GB/month. Azure Premium SSD v2 runs $0.08/GB/month. GCP SSD Persistent Disk costs $0.17/GB/month — more than double the other two (GCP Disk Pricing). For a cluster with 500 GB of persistent storage, that's $40/month on AWS or Azure versus $85/month on GCP. This is one area where GKE's pricing is notably more expensive.

Cross-AZ data transfer

Kubernetes spreads pods across availability zones for resilience. That inter-AZ traffic isn't free on AWS or GCP: both charge $0.01/GB in each direction. Azure doesn't charge for cross-AZ traffic within the same region. For a cluster sending 2 TB/month cross-AZ, that's $40/month on AWS and GCP, and $0 on Azure. Over a year, this advantage saves $480 on Azure alone.

Hidden Kubernetes costs including NAT gateways ($32/mo base + $0.045/GB), load balancers ($16-22/mo each), persistent volumes ($0.08-0.17/GB/mo), and cross-AZ transfer ($0.01/GB on AWS and GCP, free on Azure) add 30-50% on top of node compute for a typical production cluster (AWS VPC Pricing; GCP Disk Pricing, 2026 ).

We've found that most Kubernetes pricing articles compare only the control plane. Here's a real scenario: a 3-node cluster running 4 vCPU / 16 GB nodes, one load balancer, one NAT gateway, and 100 GB of persistent storage. All prices are US East on-demand rates as of March 2026 (AWS, Azure, Google).

AKS wins this scenario by roughly $108/month over EKS ($1,296/year). The free control plane and free cross-AZ transfers are the difference. GKE sits in the middle — cheaper nodes offset by pricier persistent storage. But these are on-demand rates. With GCP's automatic sustained-use discounts, the node line drops ~20%, bringing GKE's total closer to AKS.

What changes the math? Scale and utilization. At 10+ nodes, the $73 control plane fee shrinks to under 3% of total spend, making it nearly irrelevant. Node pricing and discount strategy dominate at scale. For a 20-node cluster running Savings Plans or reserved instances, the three providers converge within 10% of each other.

Organizations waste an average of 27% of their cloud spend, according to Flexera's 2025 State of the Cloud Report. Kubernetes clusters are even worse: CAST AI reports that the average cluster runs at just 13% CPU utilization. That's seven out of eight CPU cores doing nothing. Here are the highest-impact optimizations per provider.

EKS cost optimization

Start with Karpenter, AWS's open-source node provisioner. It replaces the Cluster Autoscaler and selects the cheapest instance type that fits your pending pods. Combine Karpenter with Spot Instances for non-critical workloads to save 60-90% on node costs. Use VPC endpoints for ECR, S3, and CloudWatch to bypass NAT gateway data processing fees.

Buy Compute Savings Plans for your baseline capacity. A 1-year plan typically saves 30-40% versus on-demand. Let Karpenter handle burst capacity with Spot. This two-tier approach keeps costs predictable without overcommitting.

AKS cost optimization

AKS integrates with Azure Spot VMs natively. Create a separate node pool for Spot and taint it for non-critical workloads. Use the AKS cost analysis add-on (built into the Azure portal) to see cost allocation by namespace and workload. Apply Azure Reservations to your baseline nodes for up to 72% savings.

Azure's free cross-AZ transfer is already a cost advantage. Double down by using Azure Container Registry with private endpoints to avoid NAT gateway charges for image pulls. If you run Windows containers, Azure Hybrid Benefit can save 40-55% on node costs.

GKE cost optimization

GKE Autopilot is the simplest path to right-sized clusters because you pay only for requested pod resources, not node capacity. For Standard mode, enable the vertical pod autoscaler (VPA) to right-size requests automatically. GKE's sustained-use discounts apply with no action required — just run your nodes.

For the biggest savings, use Committed Use Discounts (CUDs) on your baseline. A 1-year CUD saves up to 37%, and a 3-year CUD saves up to 57%. GKE also offers Batch API pricing for short-lived batch workloads at steep discounts, making it appealing for data processing pipelines.

Flexera's 2025 State of the Cloud Report found that organizations waste 27% of cloud spend on average. In Kubernetes specifically, CAST AI's 2025 benchmark reports average CPU utilization of just 13%, meaning seven out of eight allocated CPU cores sit idle (CAST AI, 2025 ).

Sixty-one percent of CNCF survey respondents run Kubernetes on a public cloud provider (CNCF 2024 Survey), but the right provider depends on your existing stack, team skills, and workload profile. There is no universal "cheapest" option. Here's a decision framework based on real pricing differences.

Choose EKS when

Your team already runs on AWS and uses services like RDS, ElastiCache, or SQS heavily. EKS's tight integration with IAM, VPC, and CloudWatch makes it the natural choice for AWS-native architectures. Karpenter gives EKS the most sophisticated node provisioning of any managed Kubernetes service. If you're running large-scale, heterogeneous workloads, EKS with Karpenter and Spot is hard to beat on node cost optimization.

Choose AKS when

Your organization uses Microsoft 365, Active Directory, or Azure DevOps. AKS is the cheapest starting point thanks to the free control plane and free cross-AZ data transfer. For teams running .NET or Windows containers, Azure Hybrid Benefit makes AKS significantly cheaper than EKS or GKE. If you're cost-sensitive and your workload fits the Azure ecosystem, AKS is the smart default.

Choose GKE when

You want the most managed experience with the least operational overhead. GKE Autopilot handles node provisioning, security patching, and scaling automatically. GCP's sustained-use discounts reward you without requiring commitment purchases. For teams running data-intensive workloads with BigQuery, Dataflow, or Vertex AI, GKE's integration with the Google data ecosystem is unmatched. Autopilot is especially attractive for variable workloads that are hard to forecast.

Still unsure? Start with the provider where your other services already run. Multi-cloud Kubernetes sounds appealing in theory, but the added networking complexity and duplicated tooling usually costs more than the savings from cherry-picking providers. Pick one, optimize it well, and revisit the decision when your cluster count grows past five.

The control plane price gets the headlines, but it's the smallest part of your Kubernetes bill. Nodes, NAT gateways, load balancers, and persistent volumes drive 90%+ of total cost. AKS offers the lowest sticker price for small clusters. GKE wins on operational simplicity and automatic discounts. EKS dominates for teams already deep in the AWS ecosystem.

The best way to control managed Kubernetes costs isn't choosing the cheapest provider. It's right-sizing your resource requests, using spot instances for stateless workloads, and eliminating idle capacity. We've found that optimization typically saves 40-60%, dwarfing any difference between providers.

Want to estimate your own cluster costs across providers? Try the microservices cost calculator to model your workload and compare pricing.