Best Managed Kubernetes in 2026: EKS vs GKE vs AKS vs DOKS

Pick the managed Kubernetes of the cloud that already holds your identity and data. EKS, GKE and AKS Standard all charge $0.10 per cluster-hour as of October 2026; AKS Free and DOKS charge nothing. What differs is what staying on an old version costs, how much of the node layer they run, and how pods get identity and IPs.

The numbers, side by side

List prices per cluster as of October 2026; AKS Automatic and Premium figures are East US, from the Azure Retail Prices API.

Two traps hide in that table. EKS turns extended support on by default, so a cluster that drifts past 14 months quietly moves from $0.10 to $0.60 an hour, six times the fee, which is where EKS 1.34 clusters land on December 2. And AKS Automatic, Microsoft's recommended production default, costs more than Standard.

The free control plane matters on fleets: thirty per-team clusters at $0.10 an hour is about $2,190 a month before a single node. On one large cluster it is noise next to compute.

What each hands-off mode really costs

Nodes bill at standard compute rates everywhere; the question is who manages them.

EKS Auto Mode runs Karpenter-based provisioning, load balancing, EBS storage, pod networking and node patching for you, on Bottlerocket-based nodes with a 21-day maximum lifetime. The fee is billed per second on top of EC2. In the pricing page's own example, a c6a.2xlarge at $0.306 an hour carries $0.03672 an hour of Auto Mode fee, 12% of the instance price. It replaces the self-run setup in Karpenter node provisioning patterns at scale. See our EKS Auto Mode migration notes.

GKE Autopilot charges for the CPU, memory and storage your Pods request on its default compute platform, not for nodes, which removes the gap between provisioned and used capacity. Pods that select specific hardware, such as GPUs or a machine series, switch to node-based billing. The financially backed SLA is 99.95% for Autopilot and regional control planes, 99.5% for zonal Standard.

AKS Automatic is preconfigured on the Standard tier with an uptime SLA and a pod readiness SLA. The Retail Prices API lists its hosted control plane at $0.16 an hour plus per-hour management meters by compute family, from $0.007312 for high-performance compute. Plain node pools remain for teams that want direct control.

DOKS is deliberately simple: fewer knobs, flat pricing, Basic nodes from $12 a month. For a small team running a normal web app, that is a feature.

The upgrade deadlines on the calendar

Upstream Kubernetes patches each minor version for roughly a year: as of October 2026 the supported lines are 1.37, 1.36 and 1.35, after 1.34 reached end of life on 27 October 2026. GKE upgrades you on its own through release channels (Rapid, Regular, Stable, Extended) inside a maintenance window you pick; EKS and AKS wait for you, until they don't.

From each provider's version calendar:

• EKS 1.31 extended support ends November 26, 2026. EKS then auto-upgrades the control plane with no advance notice of the exact time, and leaves managed and self-managed node groups on the old version.
• AKS 1.34 reaches community end of life in November 2026. LTS customers on Premium keep it until November 2027.
• EKS 1.34 standard support ends December 2, 2026. Clusters still on it bill $0.60 an hour from that day.
• GKE 1.34 standard support ends January 25, 2027; extended support on the Extended channel runs to November 25, 2027.
• AKS starts platform-initiated upgrades on June 1, 2027 for clusters past the 60-day grace period. Microsoft says these do not follow your maintenance window, do not stop for deprecated API usage, and can override a Pod Disruption Budget.
• AKS kubenet retires March 31, 2028. Migrate to Azure CNI Overlay before then.

Check where each cluster stands today:

aws eks describe-cluster-versions \
    --query 'clusterVersions[].[clusterVersion,status,endOfStandardSupportDate]' \
    --output table
aws eks update-cluster-config --name dev --upgrade-policy supportType=STANDARD
gcloud container clusters describe prod --location=europe-west1 \
    --format="value(currentMasterVersion,releaseChannel.channel)"
az aks show -g prod-rg -n prod --query '{version: kubernetesVersion, tier: sku.tier}'
az aks get-upgrades -g prod-rg -n prod -o table

The second command opts a dev cluster out of extended support, so EKS auto-upgrades it instead of billing six times the fee. On production, upgrade on your own schedule.

Identity and networking, the differences you feel daily

Identity. EKS, GKE and AKS all map a Kubernetes service account to a cloud identity without static keys. EKS Pod Identity avoids per-cluster OIDC providers: roles trust one principal, pods.eks.amazonaws.com, and an agent DaemonSet hands out credentials (Auto Mode clusters do not need the agent installed). It is not available for Fargate or Windows pods. GKE uses Workload Identity Federation and AKS uses Microsoft Entra Workload ID, both federating projected service account tokens, the mechanism in our workload identity and projected tokens deep dive.

Networking. EKS's VPC CNI gives each pod a VPC IP, which makes pods first-class on the network and makes subnet sizing your problem; Auto Mode adds secondary CIDR support for that reason. GKE Dataplane V2 is eBPF and Cilium based, on by default for new Autopilot clusters, with NetworkPolicy always enforced. AKS recommends Azure CNI Overlay for most clusters, which hides pod IPs behind the node and conserves VNet space; Azure CNI Pod Subnet is the flat option when other networks must reach pods directly.

The failure modes we see

• The sixfold EKS bill. Alert on endOfStandardSupportDate and look for EXTENDED_SUPPORT in aws eks describe-cluster-versions.
• Upgraded control plane, stale nodes. EKS auto-upgrades only the control plane at the end of extended support; nodes and add-ons stay behind until you move them.
• An Autopilot DaemonSet that never schedules. Security agents that need host access may not be allowed. Check vendor support before migrating.
• Pod IP exhaustion on EKS. Small subnets plus VPC CNI means Pending pods with free CPU. Plan CIDRs up front or add secondary ranges.
• Production on AKS Free. No financially backed SLA. Move to Standard before launch.

Where this breaks

List prices ignore committed-use discounts, Savings Plans and enterprise agreements, any of which can flip the node-cost comparison. AKS dates given as months are estimates, so re-check calendars before a quarterly plan. And none of this applies if you should not run Kubernetes at all, which is the argument in do you need Kubernetes.

Questions people ask

Which managed Kubernetes is cheapest?

DigitalOcean DOKS and AKS Free charge nothing for the control plane, so they are cheapest for small clusters. GKE's $74.40 monthly credit covers the fee for one zonal or Autopilot cluster per billing account. Beyond that, EKS, GKE and AKS Standard all charge $0.10 per cluster-hour as of October 2026, and node costs decide the real bill.

How long is a Kubernetes version supported on EKS, GKE and AKS?

EKS supports each minor version for 14 months of standard support plus 12 months of paid extended support, 26 months in total, then auto-upgrades the control plane. GKE publishes end dates per version, with longer support on the Extended release channel. AKS supports the latest three minor versions, roughly one year each, and Premium tier LTS adds a second year.

Is GKE better than EKS?

As Kubernetes, usually yes: GKE has stronger autoscaling, channel-based automatic upgrades and Autopilot billing on pod requests. But if your data, IAM and networking live in AWS, EKS wins, because the integration is the product.

The call we'd make

Choose by data gravity: EKS if you live in AWS, AKS if identity is Entra ID, GKE if Kubernetes itself is the platform and you want the most automation. Then put the dates above in your calendar, opt dev clusters out of EKS extended support, and plan AKS upgrades before June 2027 so Microsoft does not plan them for you. The exception is a small team on a hyperscaler out of habit, paying a complexity tax for integrations it does not use. That team should look hard at DOKS.

Sources

• AWS: Amazon EKS pricing
• AWS: Kubernetes version lifecycle on EKS
• Google Cloud: GKE pricing
• Microsoft Learn: Supported Kubernetes versions in AKS