Cloud Providers

GKE vs EKS vs AKS: Managed Kubernetes Compared

Compare Google Kubernetes Engine, Amazon EKS, and Azure Kubernetes Service on control-plane pricing, automation, networking, identity, and upgrades.

On this page
  1. What managed Kubernetes includes
  2. Quick comparison
  3. Control-plane pricing and what it leaves out
  4. Automation and hands-off modes
  5. Networking and identity
  6. Versions, upgrades, and support
  7. Scale limits
  8. Which one to choose
  9. Common misconceptions

Google Kubernetes Engine (GKE), Amazon Elastic Kubernetes Service (EKS), and Azure Kubernetes Service (AKS) all run standard, upstream Kubernetes and manage its control plane for you. The differences lie in how each one prices the control plane, how much it automates, how it handles networking and identity, and how it fits with the rest of its cloud. None of the three is the right choice for every team, and most of the work of running Kubernetes happens in your own applications and nodes, which every service bills separately.

What managed Kubernetes includes

A managed service runs the Kubernetes control plane: the API server, scheduler, and related components. The provider handles their availability and upgrades. You still manage your worker nodes or choose a serverless node option, deploy your applications, and handle monitoring, security policies, and cost.

Kubernetes was created at Google, announced in June 2014, released as version 1.0 in July 2015, and donated to the Cloud Native Computing Foundation. GKE became generally available in August 2015 as Google Container Engine. EKS and AKS became generally available in June 2018.

Quick comparison

Area GKE EKS AKS
Control-plane fee $0.10/hour per cluster; a free credit covers one zonal or Autopilot cluster per billing account $0.10/hour per cluster; extended support $0.60/hour Free tier; Standard $0.10/hour with SLA; Premium $0.60/hour
Hands-off mode Autopilot (generally available February 2021; billed per pod) Auto Mode (December 2024) Automatic (generally available 2025, check current status)
Serverless pods Autopilot EKS on Fargate Virtual nodes (Azure Container Instances)
Networking VPC-native; Dataplane V2 (eBPF) VPC CNI, where pods get VPC IP addresses Azure CNI, CNI Overlay, or Cilium
Workload identity Workload Identity Federation IRSA and Pod Identity Entra Workload ID
Autoscaling Cluster Autoscaler Karpenter (created by AWS, open source) or Cluster Autoscaler Cluster Autoscaler; node autoprovisioning uses Karpenter
Command-line tool gcloud container clusters eksctl (plus the AWS CLI) az aks

Prices and feature names change, so check each provider’s documentation before planning a budget.

Control-plane pricing and what it leaves out

The control-plane fee is the most visible cost, but it is usually the smaller one. EKS charges $0.10 per hour per cluster, which is about $73 per month for a cluster that runs all month. The breakdown of why EKS is so expensive looks at the rest of an EKS bill. GKE charges the same rate, but a free credit of $74.40 per month per billing account covers one zonal or Autopilot cluster. AKS has a Free tier with no control-plane fee, and it has no financially backed SLA. Its Standard tier costs $0.10 per hour and includes an SLA. Its Premium tier costs $0.60 per hour and adds long-term support for versions.

The phrase “AKS is free” applies only to the control plane. Worker nodes are billed separately on all three platforms. A production cluster with several nodes, load balancers, storage, and egress will cost far more than the control plane. The cloud pricing comparison explains the pricing models that apply to nodes and data transfer.

Automation and hands-off modes

All three offer modes that take over more of the node layer, which reduces the operations work.

  • GKE Autopilot manages nodes for you and bills per pod. It is generally available since February 2021. It restricts some node-level customization, such as privileged pods.
  • EKS Auto Mode was introduced in December 2024 and takes over node management in a similar way. It also has restrictions on node customization.
  • AKS Automatic was generally available in 2025. Check its current status and limits before you rely on it.

For teams that want full control, the standard modes on each platform let you choose machine types, node pools, and operating-system settings. Hands-off modes trade some of that control for less operations work.

Serverless pods are another option. EKS on Fargate, AKS virtual nodes, and GKE Autopilot each run pods without you managing nodes, but the specific limits differ.

Networking and identity

Networking is one of the areas where the three services differ most. EKS uses the VPC CNI, which gives each pod an address from your VPC. AKS offers Azure CNI, which also assigns addresses from the virtual network, along with CNI Overlay and Cilium. Azure has stated that kubenet is being retired, so check the current guidance if you use it. GKE clusters are VPC-native, and GKE’s Dataplane V2 uses eBPF for networking and policy.

For identity, each platform connects Kubernetes service accounts to the cloud’s identity system. EKS offers IAM Roles for Service Accounts (IRSA) and Pod Identity. AKS uses Microsoft Entra Workload ID. GKE uses Workload Identity Federation. Each approach has its own setup steps, so decide on the identity model early.

Versions, upgrades, and support

Every managed service requires you to upgrade Kubernetes versions. Approximate support windows are about 14 months of standard support on EKS and GKE, with EKS offering an extended support period for an additional 12 months at a higher control-plane rate. AKS supports each version for 12 months, and its Premium tier supports selected versions for up to two years.

A managed control plane does not test your applications for version changes. Plan a staging cluster, read the deprecation notices for each version, and test your workloads before you upgrade production. Managed upgrades reduce the work, but they do not remove it.

Scale limits

Scale limits change, so check current figures before designing for a large cluster. Published limits at the time of writing were about 5,000 nodes for AKS, 15,000 nodes for standard GKE clusters, with a higher limit of 65,000 announced in November 2024. EKS was announced in July 2025 to support up to 100,000 nodes, which should be verified against current documentation.

Most teams will not approach these limits. Multi-cluster management matters more for many organizations, and GKE has strong fleet tooling for that purpose.

Which one to choose

  • Choose EKS if your organization already standardizes on AWS networking, IAM, and compute, and you want Kubernetes close to the rest of your AWS services.
  • Choose AKS if your organization relies on Microsoft identity and governance, or if you want the Free tier to keep control-plane costs at zero for development clusters.
  • Choose GKE if you want the most automated Kubernetes experience, strong AI and machine learning support, or multi-cluster fleet features.

These are starting points. Your existing cloud, team skills, and compliance rules usually matter more than small differences in features. Run a short proof of concept with your own application before committing.

Common misconceptions

  • “AKS is free” applies only to the Free tier control plane. Nodes still cost money.
  • A managed control plane does not test your workloads. You still need to test each upgrade.
  • Autopilot and Auto Mode restrict some node settings. Privileged pods and some node customizations may not be allowed.
  • Upstream Kubernetes is the same everywhere. The differences are in networking, identity, upgrades, and pricing, not the core API.

For the wider platform context, read the AWS vs Azure vs Google Cloud comparison. If you are considering serverless containers instead of Kubernetes, the serverless comparison covers that option. For each provider’s overview, see what Google Cloud Platform is used for and what Microsoft Azure is used for.

The complete guide to cloud providers

AWS vs Azure vs Google Cloud: A Full Comparison
Read the guide

Keep exploring

Guides from other topics

All topics →