What actually changed
In standard ARO, the control plane (API server, etcd and related components) runs on virtual machines in your own Azure subscription. In the hosted model, the control plane runs in a Microsoft-managed Azure subscription, and only the worker nodes live in your subscription and virtual network. The design is based on the HyperShift project. Red Hat's preview announcement described it as a new deployment model that helps optimize costs, increase flexibility and improve efficiency.
The Microsoft Learn comparison page gives these differences:
| Area | Standard architecture | Hosted control planes |
|---|---|---|
| Control plane location | Dedicated VMs in your subscription | Managed environment outside your subscription |
| Minimum footprint | 3 control plane nodes + 3 workers | 2 worker nodes, no control plane nodes in your subscription |
| Infrastructure nodes | Dedicated nodes for ingress, registry and monitoring | None; those components run on worker nodes |
| Upgrades | Control plane and workers upgraded together | Control plane and each node pool upgraded separately |
| Provisioning time (Microsoft's estimate) | About 45 minutes | About 15 to 20 minutes |
| Billing | All nodes, including control plane | Worker nodes only (minimum two) |
The Learn page also says the hosted control plane talks to workers over a dedicated VNet integration subnet. That subnet is delegated to Microsoft.RedHatOpenShift/hcpOpenShiftClusters. Architects should plan for it in their network design.
Cost claims: read the fine print
Microsoft says preview customers saw "significant VM cost reduction" but gives no figure. Red Hat's preview post claimed savings up to 4x on average, based on a Red Hat internal study. That is a vendor figure, not a guaranteed outcome.
Going from six required nodes to two does cut compute. The actual savings depend on your VM sizes, your worker count and the ARO license component charged per application node. Without a worker-count baseline, treat any percentage as marketing until you've run your own numbers. Because dedicated infrastructure nodes are gone, ingress, registry and monitoring now compete with your applications for worker capacity. Size your workers with that in mind.
Upgrades: two lifecycles
Separate lifecycles are probably the most useful operational change.
- Control plane patches (z-stream): Microsoft and Red Hat site reliability engineers apply these automatically.
- Control plane minor versions (y-stream): You start these, unless a minor version reaches end of life.
- Node pools: They can run up to two minor versions behind the control plane. They can also be provisioned on a specific OpenShift version.
Microsoft's suggested pattern is to create a new node pool on the newer version, move workloads over, then retire the old pool. That lets each application team upgrade on its own schedule. The announcement doesn't spell out deadlines or the exact end-of-life procedure, so check the service lifecycle documentation before you rely on it.
Management and security features
Day-two tasks run through Azure Resource Manager. You can create and resize node pools, set autoscaling limits and connect Microsoft Entra ID. You can do this from the Azure portal, Azure CLI, ARM templates or Bicep. Existing Azure governance and infrastructure-as-code habits carry over.
The announced security features are:
- Customer-managed keys: OS disks are encrypted with them, and managed HSM-backed key vaults are supported at GA.
- Identity provider: You can configure your own for administrator and user sign-in.
- Least privilege: Managed identities with separate role assignments cover the OpenShift components. Workload identity lets applications use token-based authentication instead of usernames and passwords.
- Sandboxed containers: Red Hat OpenShift sandboxed containers add an extra isolation boundary for sensitive workloads.
These are capabilities, not a compliance certification. Red Hat also says the architecture reduces the potential attack surface of the environment by restricting direct access to the cluster. You still need to check identity, key management, networking and isolation against your own regulatory requirements. Your auditors may have views on a control plane that sits outside your tenant.
Regions and ecosystem
Microsoft lists eight GA regions:
- UK South
- Canada Central
- Australia East
- Switzerland North
- Brazil South
- Central India
- US East 2
- West Europe
Sweden Central and East Asia come after GA, followed by more regions worldwide. Those two are planned, not live.
Other GA items:
- Red Hat Advanced Cluster Management 5: It can create, upgrade, modify and delete hosted-control-plane clusters. It can also apply governance policies from one console.
- Isovalent Networking for Kubernetes: It is certified as a supported CNI. Isovalent says the certification covers INK 1.19 and later. It also says ARO with hosted control planes lets you choose an alternative CNI at cluster creation.
- Red Hat OpenShift 5 support: Microsoft's announcement says the model supports it. It also describes OpenShift 5 features, including an MCP server and gateway for connecting AI agents, OpenShift Lightspeed and post-quantum cryptography protections. Those belong to OpenShift 5, not to the hosted architecture itself.
Documentation gaps
The quickstart that Microsoft's announcement points to is still labeled "preview". The Learn comparison page doesn't carry a version-specific matrix. Don't assume every preview-era limitation, or every feature, matches the GA state. Check the current docs for your region and version.
Before you migrate
- Confirm your target region is on the GA list.
- Check the OpenShift version and node-pool requirements.
- Plan the delegated VNet integration subnet.
- Model costs with at least two workers, plus any extra worker capacity needed for platform components.
- Decide whether a control plane outside your subscription fits your operating and regulatory model.
- Confirm that any third-party networking you depend on is supported.
Microsoft's materials don't describe in-place migration of existing standard clusters. Treat this as a new-cluster option for now.
Hosted control planes give teams a leaner cluster and faster provisioning. They also give you more flexible upgrades and Azure-native management. The trade-off is less visibility into the control plane. Most teams should pilot it rather than rush to move production onto it.
References
- (Launched) Generally Available: Azure Red Hat OpenShift with hosted control planes Azure Updates · 2026-11-17T20:00:32Z
- Azure Red Hat OpenShift azure.microsoft.com
- Isovalent Networking for Kubernetes Certified for Microsoft Azure Red Hat OpenShift isovalent.com