Anthos vs GKE: When Each Wins (and When You're Overpaying)

If you’re reading a “Anthos vs GKE” article written before late 2025, discard it. Not because the authors were wrong at the time — but because the product they were writing about no longer exists in any recognizable form.

Here’s the compressed version: Google rebranded Anthos to GKE Enterprise in November 2023. Then, in September 2025, Google dissolved the Enterprise tier entirely. Their own documentation now states: “GKE is a single offering, without different editions or tiers. The features that were part of GKE Enterprise have become part of the standard Google Kubernetes Engine (GKE) offering, or are products or features that you can use in addition to GKE.”

What this means in practice: the “Anthos vs GKE” framing is now a category error. There is one GKE. Most of what you used to pay for as Anthos — Fleets, Config Sync, Policy Controller, Connect Gateway, Fleet observability — is now included in standard GKE at no extra cost. The genuinely expensive pieces, Cloud Service Mesh, Multi-Cloud cluster management, Backup for GKE, Extended Support, and Multi-cluster Gateway, split off into standalone à-la-carte SKUs.

The problem is that most teams haven’t caught up. Many are still paying for configurations that made sense under the old bundled pricing model. The AI Overview Google surfaces for this query still cites a 2022 article that describes Anthos clusters as “extensions of GKE.” That was true four years ago. The billing reality in 2026 is different, and the gap is costing mid-market teams real money.

This article covers what’s free now, what the current SKU pricing actually is, and three configurations where teams are routinely overpaying.

What “Anthos” Actually Means in 2026

In 2026, “Anthos” refers to a set of optional paid SKUs layered on top of standard GKE — not a separate tier, edition, or product. Most former Anthos capabilities are now free in base GKE. The only features that still carry a separate charge are Cloud Service Mesh, Multi-Cloud cluster management, Backup for GKE, Extended Support, and Multi-cluster Gateway.

The timeline matters for understanding your bill. From 2019 to late 2023, Anthos was a premium subscription on top of GKE — originally priced at $10,000 per month per 100-vCPU block, sold in 100-vCPU increments on annual contracts. That’s the number many vendors and comparison sites still quote. It’s completely stale.

When Google rebranded to GKE Enterprise in November 2023, they moved to hourly per-vCPU billing at $0.00822/vCPU/hour for “Enterprise vCPUs under management.” Less brutal than the original pricing, but still a significant lift for single-cloud teams who didn’t need the multi-cluster features.

The September 2025 consolidation changed the math again. The Enterprise vCPU management fee was eliminated entirely for standard GKE use. Features that previously required the Enterprise tier — Config Sync for GitOps cluster config, Policy Controller for OPA Gatekeeper-style admission control, Fleet APIs for grouping clusters, Connect Gateway for authenticating to non-GCP clusters — all moved into base GKE.

What remains as paid add-ons is a shorter, more defensible list: the capabilities that genuinely require infrastructure Google runs on your behalf (managed service mesh control plane, off-GCP cluster management agents, namespace-level backup) or features with significant operational costs (Extended Support for older Kubernetes versions, Multi-cluster Gateway for cross-cluster load balancing).

This is a better model than the old bundle. The problem is that teams who enabled the Anthos API or the old Enterprise tier — and haven’t explicitly disabled it — may still be on the hook for line items that survived the consolidation.

Live 2026 Pricing — The Actual SKUs

Base GKE costs $0.10/cluster/hour (roughly $72/month) for every cluster, every mode. Beyond that, you pay only for the add-on SKUs you explicitly enable. Here are the current rates from cloud.google.com/kubernetes-engine/pricing:

SKU	Rate	Monthly estimate
Cluster management fee	$0.10/cluster/hr	~$72/cluster
Extended Support (older K8s versions)	+$0.50/cluster/hr on top of base	+~$432/cluster/mo
Cloud Service Mesh (managed)	~$0.0066–$0.10/vCPU/hr (tier-dependent)	~$70–$1,000+ at 100 vCPU
GKE Multi-Cloud — AWS/Azure clusters	$0.00822/vCPU/hr	~$6/vCPU/mo
GKE Multi-Cloud — Attached Clusters	$0.10/vCPU/hr	~$73/vCPU/mo
GDC — vSphere or Bare Metal on-prem	$0.03288/vCPU/hr	~$24/vCPU/mo
Multi-cluster Gateway / Ingress	$3/backend Pod/mo ($0.0041/Pod/hr)	$60 at 20 Pods
Backup for GKE	$9/namespace/mo + $0.045/GiB/mo	varies

A few numbers worth flagging explicitly.

Extended Support adds $0.50/cluster/hour on top of the base $0.10 fee, bringing the total to $0.60/cluster/hour — roughly $432/month per cluster. This applies only if you’re running a Kubernetes version past its standard support window. Teams who haven’t kept up with K8s upgrades often don’t realize they’re paying this surcharge.

Attached Clusters at $0.10/vCPU/hour is the sleeper charge in the Multi-Cloud lineup. Run-by-Google clusters on AWS or Azure cost $0.00822/vCPU/hour. Bring-your-own CNCF-compliant clusters (your own EKS, AKS, or on-prem cluster connected via the GKE API) cost $0.10/vCPU/hour — 12 times the rate. Many teams discovered this when migrating an existing EKS cluster under GKE management and got a bill that was an order of magnitude larger than expected.

Cloud Service Mesh pricing varies by tier and configuration. Google’s pricing page lists rates ranging from approximately $0.0066/vCPU/hour at the low end to $0.10/vCPU/hour for the managed control plane tier. At 100 vCPUs with the managed control plane, that’s $7,300/month — before compute, before egress.

These numbers come directly from cloud.google.com/kubernetes-engine/pricing. Verify them before budgeting; Google updates SKU pricing independently of product announcements.

The 3 Overpay Configurations

Three configurations account for the majority of Anthos-era overpay we see in 2026: single-cluster teams who never disabled the Enterprise API, teams running Cloud Service Mesh on flat microservice networks that don’t need it, and hybrid teams paying for multi-cluster management when open-source tooling would be cheaper.

Configuration A: Single-Cluster, Single-Cloud Teams Still on the Legacy API

This is the most common scenario. A team is running a single GKE cluster in one region, entirely within Google Cloud. Someone on the platform team — or a consultant, or a sales rep — enabled the GKE Enterprise tier or the Anthos API when the cluster was set up, because “Enterprise” sounded like the right choice for production.

Before September 2025, this meant paying the Enterprise vCPU management fee: 60 vCPUs × $0.00822/hr × 730 hours = roughly $360/month for nothing useful. On top of that, many of these teams also had Cloud Service Mesh and Multi-cluster Ingress enabled by default, adding another $300–$600/month. Annual overpay: $8,500 to $15,000 for capabilities they aren’t using.

Post-September 2025, the Enterprise vCPU fee was eliminated — but the Anthos API itself, if still enabled, can keep certain billable components active. Teams who haven’t audited their cluster configuration may still be paying for Extended Support they don’t need, or for Cloud Service Mesh that was enabled alongside the Enterprise tier and never turned off.

The audit command is simple:

gcloud container clusters describe <cluster-name> \
  --location=<location> \
  --format='value(enableK8sBetaApis,addonsConfig)'

Check your billing export for line items containing anthos, service-mesh, or multi-cluster-ingress. If you’re a single-cluster, single-cloud team and you see any of those, disable them. The fix is documented at cloud.google.com/kubernetes-engine/enterprise/docs/setup/disable-anthos. Fleets, Config Sync, and Policy Controller will continue to work — they’re now in base GKE and don’t require the Anthos API.

If your gcloud container clusters describe output still shows the old Anthos API enabled and you’re a single-cluster, single-cloud team, you’re paying roughly $432/month for Extended Support you may not even be using. Turn it off.

Configuration B: Cloud Service Mesh on a Flat Three-Service Network

This one is harder to blame on anyone in particular. Service meshes became fashionable, “zero-trust” became a mandate from security, and Cloud Service Mesh was the obvious choice if you were already running on GKE. So teams turned it on.

The problem is that Cloud Service Mesh earns its cost on complex networks: 25+ services, cross-cluster traffic routing, canary deployments that need traffic-weight control, or zero-trust requirements that extend across multiple VPCs. On a three-service application where all services are in the same VPC talking over ClusterIP, you’re paying for an F1 car to drive to the grocery store.

At the managed control plane tier, Cloud Service Mesh runs roughly $0.10/vCPU/hour. On a 100-vCPU cluster, that’s approximately $7,300/month — $87,600/year — for capabilities you’re likely replicating with tools you already have. GKE Dataplane V2 (Cilium-based, free in all GKE clusters) handles network policy enforcement. GKE Workload Identity (free) handles service-to-service authentication. Cloud Logging and Cloud Trace (already in your GCP bill) cover the observability layer.

If you have fewer than 10 services, no cross-cluster traffic routing, and no compliance requirement that explicitly mandates a managed service mesh, disable Cloud Service Mesh. Open-source Linkerd or Istio Ambient mode covers mTLS between services at zero licensing cost. The operational overhead of running OSS Istio is real, but at $87,600/year in avoided SKU costs, it’s worth the 0.1–0.2 FTE of platform engineering time.

Configuration C: Hybrid Teams Paying for Multi-Cluster Management on One Small On-Prem Cluster

This is the “single pane of glass” pitch that Google’s sales team made effectively for years. The scenario: 85% of workloads on GCP, 10% on a VMware on-prem cluster kept for compliance reasons, 5% in an AWS dev account. Anthos (now GKE Multi-Cloud + GDC) promised unified management across all three.

The cost at 200 total vCPUs:

• GDC (vSphere) on-prem: 20 vCPU × $0.03288/hr × 730 hr = $480/month
• GKE Multi-Cloud (AWS): 10 vCPU × $0.00822/hr × 730 hr = $60/month
• Cloud Service Mesh across the fleet, Config Sync managed: add another $400–$600/month
• Total: $700–$1,500/month for the orchestration layer alone, roughly $10,000–$18,000/year

The open-source alternative that covers 90% of the same use case:

• Argo CD (free) for GitOps across all three environments — replaces Config Sync
• Open-source Istio or Linkerd (free) for the on-prem cluster — replaces Cloud Service Mesh
• Velero (free) for backup on the on-prem cluster — replaces Backup for GKE there
• Terraform or Crossplane for cross-environment resource management

The engineering cost to operate this stack is roughly 0.25–0.5 FTE annually, call it $50,000–$100,000 depending on your market. The math works in the OSS stack’s favor once you’re below roughly 40 on-prem vCPUs or 60 multi-cloud vCPUs. Above 100 on-prem vCPUs with a team that can’t staff a dedicated platform engineer, the managed SKUs start to make economic sense.

The critical number to check is the Attached Clusters rate. If you connected an existing on-prem cluster or EKS cluster using GKE Attached Clusters, you’re paying $0.10/vCPU/hour — not the $0.00822/vCPU/hour rate for Google-run clusters. On 40 on-prem vCPUs, that’s $2,920/month versus $240/month. That difference almost always justifies the move to Argo CD + Istio OSS.

What Native GKE Standard and Autopilot Give You for Free

As of September 2025, standard GKE includes nearly everything that previously required an Anthos or GKE Enterprise subscription. The following capabilities cost nothing beyond the base $0.10/cluster/hour management fee:

Feature	What it does
Fleets	Group and manage multiple clusters as a unit
Teams (RBAC)	Fleet-scoped role bindings across clusters
Config Sync	GitOps for cluster configuration (was Anthos-only)
Policy Controller	OPA Gatekeeper admission control (was Anthos-only)
Fleet dashboard	Multi-cluster observability in Cloud Console
Connect Gateway	Auth-broker access to non-GCP clusters
Multi-cluster Services (MCS)	Service discovery across clusters
Workload Identity Federation	Pod-to-GCP-service authentication
GKE Dataplane V2 (Cilium)	eBPF-based networking, network policy enforcement
Config Connector	Manage GCP resources via Kubernetes CRDs
GKE Cost Allocation	Namespace and label-level cost attribution
Security Posture dashboard	Vulnerability scanning, compliance views
Binary Authorization	Container image signing and policy enforcement
GKE Sandbox (gVisor)	Workload isolation for untrusted code
Financial-backed SLA	99.95% regional / 99.5% zonal uptime guarantee

GKE Autopilot is worth calling out specifically for mid-market teams. Autopilot removes node management entirely — you pay per pod rather than per node, and the control plane is included in the pod price. In us-central1, Autopilot costs approximately $0.0445/vCPU/hour and $0.0049/GiB/hour for memory. For smaller teams running bursty workloads, Autopilot consistently comes out cheaper than Standard because you’re not paying for idle node headroom.

For teams coming off an Anthos or Enterprise contract and re-evaluating their cluster setup, Autopilot is almost always the right starting point. Add paid SKUs only when you hit a specific requirement that Autopilot’s included features don’t address.

When the Paid SKUs Actually Earn Their Cost

The paid SKUs justify their cost in three specific scenarios: regulated workloads with hard RPO requirements (Backup for GKE), genuinely complex multi-service networks with traffic-splitting needs (Cloud Service Mesh), and teams with more than 100 vCPUs outside GCP and no capacity to run a platform engineering function (Multi-Cloud managed clusters).

Cloud Service Mesh makes sense when you have 25 or more services that need traffic-shifting between versions, cross-cluster load balancing, or a compliance requirement (FedRAMP, SOC 2 Type II with zero-trust scope) that mandates a managed, audited service mesh with centralized policy. Below that threshold, open-source Istio or Linkerd handles the job. The managed control plane is worth paying for when your platform team is small and the operational burden of running OSS Istio at scale would cost more in engineering time than the SKU itself.

Backup for GKE at $9/namespace/month plus $0.045/GiB/month is defensible for HIPAA or PCI workloads with an RPO under 24 hours. It provides application-consistent backup at the namespace level with restore testing built in. For teams without compliance-driven RPO requirements, Velero with a GCS backend is free and handles the same workflow.

GKE Multi-Cloud for AWS or Azure (the Google-run cluster variant at $0.00822/vCPU/hour, not Attached Clusters) makes sense when you genuinely need to run production workloads in multiple clouds and your team doesn’t want to operate separate EKS or AKS control planes. The unified API and consistent observability model has real value at scale. Be honest about whether you’re in that category or whether “multi-cloud” is aspirational. Most mid-market teams are running one significant cloud and a legacy on-prem environment — that’s a hybrid story, not a multi-cloud one, and the OSS toolchain handles it more cheaply.

A word on Attached Clusters: the $0.10/vCPU/hour rate for bring-your-own clusters is pricing that should give you pause. It means Google is charging you $73/vCPU/month to manage a cluster you already operate. At any meaningful scale — more than 20 attached vCPUs — Argo CD plus your existing cluster management tooling will be cheaper within the first year.

The Mid-Market Decision: Two Scenarios and an Audit Checklist

For most mid-market teams, the answer is standard GKE (Autopilot preferred) with zero paid add-on SKUs. The exceptions are narrow and specific: regulated workloads needing guaranteed restore capability, or genuine multi-region production presence outside GCP.

Scenario 1: 50–300 employee SaaS company, single cloud, one or two clusters. Use Autopilot. Full stop. You get Fleets, Config Sync, Policy Controller, Workload Identity, and Security Posture for free. No Cloud Service Mesh, no Backup for GKE unless you have a regulatory mandate, no Multi-Cloud unless you actually run production workloads outside GCP. Your annual Kubernetes infrastructure bill at this scale should be $1,500–$3,000/year in management fees plus compute. If it’s significantly higher, something got enabled that shouldn’t have been.

Scenario 2: Mid-market regulated company (healthcare, fintech), multi-region, with one legacy on-prem cluster. GKE Standard (not Autopilot — you need more control over node configuration for compliance tooling) in two regions. Add Backup for GKE for the namespaces with PHI or PCI scope. Evaluate Cloud Service Mesh only if your service count crosses 25. For the on-prem cluster, run Argo CD for GitOps rather than paying Attached Cluster rates. Total add-on spend: $200–$500/month — not $2,000+.

The three-question audit before enabling any paid SKU:

1. Do we have 25 or more services that need traffic-shifting, canary routing, or cross-cluster load balancing? If no, skip Cloud Service Mesh.
2. Do we run real production workloads outside GCP — not dev clusters, not legacy infrastructure we’re actively migrating off? If no, skip Multi-Cloud SKUs.
3. Do we have an RPO under 24 hours mandated by a compliance framework? If no, skip Backup for GKE and use Velero on GCS.

If you answer yes to one of these, add that one SKU. Most mid-market teams answer yes to zero.

Find a Google Cloud Partner Who Knows the 2026 Billing Reality

The partner conversation matters here because most Google Cloud consultants still quote Anthos pricing from 2022 or 2023. They’re using stale rate cards and, more importantly, they’re recommending architectural patterns that made sense under the old bundled model and don’t hold up under the current à-la-carte SKU structure.

A partner who understands the September 2025 consolidation will tell you: start with Autopilot, enable Config Sync and Policy Controller at no extra cost, and add paid SKUs only when you can point to a specific requirement that the free tier doesn’t satisfy. That’s different from the conversation that results in a $2,000/month Anthos invoice for a team running three microservices on a single cluster.

For mid-market teams navigating hybrid cloud architecture decisions, the choice between managed SKUs and open-source tooling is the same question — what does the operational burden cost versus what does the managed service cost — applied to a specific set of line items. The cloud cost optimization framework applies directly here: audit what’s enabled, calculate the delta against alternatives, and default to the simpler option unless the complexity is justified.

If you’re evaluating Google Cloud managed services providers, the ability to navigate this specific billing shift — from bundled Anthos to à-la-carte SKUs — is a reasonable proxy for whether a partner is current on GCP pricing or still working off a 2022 playbook.

See our ranked Google Cloud consulting partners — vetted and ranked on criteria that include pricing transparency and current platform knowledge, not just certification counts.