Self-managed Red Hat OpenShift is sold as a yearly subscription for the compute nodes of your clusters. You license either every core-pair, which is 2 physical cores or 4 vCPUs, or every bare-metal server, whatever its socket and core count. GPUs and other accelerators need their own subscription per device. Managed OpenShift in the cloud is billed by the hour instead: ROSA on AWS and ARO on Azure both charge $0.171 per hour for every 4 worker vCPUs, on top of the cloud infrastructure.
Red Hat does not publish list prices for the self-managed editions, but resellers do. This post explains how Red Hat counts subscriptions, what each edition costs, and what a typical cluster adds up to.
OpenShift editions and list prices
Red Hat sells four self-managed editions. Each includes the one below it:
- OpenShift Virtualization Engine runs virtual machines only, on bare metal, with no support for your own application containers.
- OpenShift Kubernetes Engine adds Red Hat Enterprise Linux CoreOS and enterprise Kubernetes for containers.
- OpenShift Container Platform adds the developer console, Serverless, Service Mesh, Pipelines, and GitOps.
- OpenShift Platform Plus adds Advanced Cluster Management, Advanced Cluster Security, OpenShift Data Foundation Essentials, and the Quay registry.
A German reseller lists the container editions at these prices for a one-year subscription, net of VAT:
| Edition | Core-pair, Standard | Core-pair, Premium | Bare-metal server, Standard | Bare-metal server, Premium |
|---|---|---|---|---|
| OpenShift Kubernetes Engine | €639 | €967 | €7,667 | €11,616 |
| OpenShift Container Platform | €2,614 | €3,872 | €23,232 | €34,848 |
| OpenShift Platform Plus | €3,243 | €4,840 | €29,185 | €43,560 |
Standard support covers business hours (8x5), and Premium covers 24x7. For Platform Plus, the reseller’s three-year terms cost exactly three times the yearly price, so the multi-year commitment brings no list discount. Large customers negotiate below list, but these numbers are the starting point for any quote.
How Red Hat counts subscriptions
The rules in Red Hat’s subscription guide decide most of the bill, so they are worth knowing in detail.
Only compute nodes count. Nodes that run your applications need subscriptions. Control plane nodes and infrastructure nodes, such as the ones that run the ingress routers, are included, as long as no application Pods run on them. Compact three-node clusters and single-node OpenShift are the exception: there, every core on every node counts.
Core-pairs follow vCPUs. On virtual machines and in the cloud, one core-pair covers 4 vCPUs, because Red Hat assumes hyperthreading is on. Core-pairs are counted across all your clusters, so 100 core-pairs can cover several clusters at once.
Bare-metal subscriptions follow servers. One bare-metal node subscription covers one physical server, regardless of how many sockets it has or how many cores its CPUs have. A subscription cannot be split across servers.
Accelerators are extra. Every physical GPU, TPU, or other accelerator used for compute needs its own AI accelerator subscription, with the same support level as the node it sits in. An 8-GPU server therefore needs eight accelerator subscriptions on top of its CPU subscription.
The operating system is included. Red Hat Enterprise Linux CoreOS for control plane, infrastructure, and compute nodes is part of every container edition. Kubernetes Engine, Container Platform, and Platform Plus also include RHEL guest entitlements for virtual machines on OpenShift Virtualization. OpenShift Virtualization Engine does not, so its RHEL guests need their own subscriptions.
One edition per cluster. All nodes in a cluster must use the same edition, although core-pair and bare-metal subscriptions can be mixed within it.
Disaster recovery counts when it is hot. A standby cluster that runs and replicates data continuously needs full subscriptions. Warm and cold standby clusters do not, until you fail over to them.
Worked example: a cluster on virtual machines
Take a production cluster with ten worker VMs of 16 vCPUs each. That is 160 vCPUs, or 40 core-pairs:
| Edition | Standard support | Premium support |
|---|---|---|
| OpenShift Kubernetes Engine | €25,560 | €38,680 |
| OpenShift Container Platform | €104,560 | €154,880 |
| OpenShift Platform Plus | €129,720 | €193,600 |
Those figures are per year, at list price, before the VMs themselves, the control plane and infrastructure nodes you still run, and the people who install and upgrade the platform. The same 160 vCPUs on Cloudfleet Pro cost about €8,900 a year: €69 a month with 24 vCPUs included, plus €4.95 per additional vCPU per month, with the managed control plane, upgrades, and support included.
Worked example: OpenShift on-premises on bare metal
On dense servers, bare-metal subscriptions are usually cheaper than core-pairs. Take six two-socket servers with 64 physical cores each:
- Core-pairs: 64 cores are 32 core-pairs per server, so 192 core-pairs in total, or about €502,000 a year for OpenShift Container Platform with Standard support.
- Bare-metal subscriptions: one subscription per server, so 6 subscriptions, or about €139,000 a year for the same edition and support level.
Add GPUs and the accelerator subscriptions follow: six servers with eight GPUs each need 48 accelerator subscriptions on top. Hardware and power come on top, including at least three control plane machines per cluster in your data center, unless you run hosted control planes on another OpenShift cluster.
ROSA pricing on AWS
Red Hat OpenShift Service on AWS (ROSA) is billed through AWS. ROSA with hosted control planes has two service fees:
- Worker nodes: $0.171 per hour for every 4 vCPUs, which is $1,500 per 4 vCPUs a year on demand, $1,000 with a one-year contract, and $667 with a three-year contract.
- Cluster fee: $0.25 per hour per cluster, or $2,190 a year.
EC2 instances, EBS volumes, and networking are billed separately. AWS’s own example of a nine-node m5.xlarge cluster comes to $11,190 a year in ROSA fees plus $12,366 in AWS infrastructure. For the ten 16 vCPU workers from the example above, the ROSA fees alone are about $62,000 a year on demand, before the EC2 bill.
ROSA classic has no cluster fee, but you pay for the EC2 instances of its control plane and infrastructure nodes. In AWS’s example, the same nine-node cluster costs $32,522 a year on classic against $23,556 with hosted control planes.
ARO pricing on Azure
Azure Red Hat OpenShift (ARO) charges an OpenShift license of $0.171 per hour per 4 worker vCPUs, the same rate for every VM type, on top of standard Linux VM prices. Control plane nodes pay no license fee. On clusters with the standard architecture, the license can be reserved for one or three years, and the VMs can use Azure reservations or savings plans. ARO with hosted control planes adds a separate cluster management fee.
Red Hat’s own pricing page quotes reserved managed OpenShift from $0.076 per hour per 4 vCPUs, based on a three-year contract. OpenShift Dedicated, the Red Hat-operated service on AWS and Google Cloud, is quoted by Red Hat.
How to lower your OpenShift bill
The largest savings come from licensing fewer cores. Check that no application Pods run on infrastructure or control plane nodes, since that turns them into billable compute nodes. Right-size worker VMs, because every idle vCPU in a worker counts. On dense on-premises servers, compare bare-metal subscriptions with core-pairs. And if your teams do not use the developer console, Pipelines, or Service Mesh, OpenShift Kubernetes Engine costs a quarter of OpenShift Container Platform.
The bigger question is which workloads need OpenShift at all. Most applications use standard Deployments, Services, and Helm charts, and they run on any conformant Kubernetes. OpenShift vs Kubernetes explains what you get for the subscription, and migrating from OpenShift to Kubernetes covers the few OpenShift-only APIs that need converting.
Cloudfleet prices managed Kubernetes per vCPU with no subscriptions, no per-core licensing, and no separate support contract, and one cluster can span your clouds and your own servers. See pricing for the plans, the Cloudfleet vs OpenShift comparison for the features, and the OpenShift alternative page for the Migration Acceleration Program, which offsets the cost of running both platforms during the move.
Frequently asked questions
How much does OpenShift cost?
At reseller list prices, OpenShift Container Platform costs €2,614 per core-pair a year with Standard support and €3,872 with Premium. A cluster with 160 worker vCPUs costs about €105,000 to €155,000 a year in subscriptions. Managed OpenShift on AWS or Azure costs $0.171 per hour per 4 worker vCPUs, plus infrastructure.
Is OpenShift licensed per core?
Yes. Self-managed OpenShift is licensed per core-pair, which is 2 physical cores or 4 vCPUs, or per bare-metal server, regardless of its socket and core count. Only compute nodes count.
Do OpenShift control plane nodes need a subscription?
No, as long as they run only OpenShift’s own components. The same applies to infrastructure nodes. In compact three-node clusters and single-node OpenShift, every core counts, because applications run on the control plane nodes.
Does OpenShift include RHEL?
Yes, for the nodes. Red Hat Enterprise Linux CoreOS for control plane, infrastructure, and compute nodes is included in the container editions, and so are RHEL guest entitlements for VMs on OpenShift Virtualization. Only OpenShift Virtualization Engine needs separate RHEL subscriptions for its guests.
How much does ROSA cost?
ROSA with hosted control planes costs $0.171 per hour for every 4 worker vCPUs, about $1,500 per 4 vCPUs a year on demand, plus $0.25 per hour per cluster. EC2, storage, and networking are billed separately by AWS.
How much does OpenShift cost on-premises?
The subscription is the same as in the cloud, per core-pair or per bare-metal server. One bare-metal server, whatever its core count, costs €23,232 a year for OpenShift Container Platform with Standard support at list price. Add accelerator subscriptions for GPUs, and your own hardware for control plane and infrastructure nodes.
