OpenShift was built to deliver applications to your own developers. Selling GPU capacity is a different business. vCluster gives every customer vanilla, CNCF-certified Kubernetes on your hardware, with Private Nodes as the production default for dedicated GPU nodes.
In our analysis, the platform model assumes your tenants are colleagues. GPU clouds sell to strangers.
Hosted control planes run on an OpenShift management cluster, and each hosted cluster is a full OpenShift cluster. Your customers inherit a platform when what they asked for was Kubernetes.
Subscriptions attach to the worker nodes you are trying to monetize. That puts a floor under every GPU hour before a customer runs anything.
Integrated registry, Source-to-Image, pipelines, developer console. Application delivery machinery a GPU cloud carries and never charges for.
vCluster virtualizes the Kubernetes control plane, so every customer gets a real, CNCF-certified cluster with their own API server, etcd, and RBAC, running as a lightweight pod and ready in seconds. Private Nodes, the production default, assign dedicated GPU nodes with per-tenant CNI and storage over an encrypted WireGuard VPN. vMetal takes the stack down to the rack. Proven across 100K+ GPU nodes and 50+ GPU clouds.
What a GPU cloud needs from the layer underneath it, without the application delivery surface on top.
Each customer gets their own API server, etcd, scheduler, and RBAC running as a lightweight pod on your control plane cluster. Marginal cost per tenant is near zero, which is what makes an entry tier worth selling.
Every tenant cluster is fully CNCF-conformant with 100% API compatibility. Customers arriving from EKS or GKE keep the tooling they already use, install their own CRDs and operators, and have nothing new to learn.
Private Nodes, the production default, join dedicated physical nodes into each tenant cluster over an encrypted WireGuard VPN. Every tenant runs their own CNI, so one customer can use SR-IOV with Multus for RDMA while another runs Calico with eBPF.
vCluster Standalone is a CNCF-certified control plane that runs as a single binary directly on Linux, with no k3s, kubeadm, or other base distribution underneath. A lean Kubernetes layer for hardware you sell by the hour.
Run every tenant cluster from a single UI, CLI, and API, with SSO, quotas, and RBAC across the fleet. Templates codify each tier, so a customer picking a plan lands on exactly the configuration you defined.
This isn’t a side project. Behind every vCluster deployment is 5+ years of deep K8s engineering, security hardening, and battle-tested infrastructure work at massive scale.
Talk to our team about your stack
Deploy vCluster on your infra in minutes
Go live with a hyperscaler-grade tenant experience in days
OpenShift was designed to deliver applications to an organization's own developers. Selling GPU capacity is a different business: your tenants are external customers who want a Kubernetes cluster and their own tooling rather than a platform. In our analysis, that mismatch is what starts the search. vCluster gives each customer vanilla, CNCF-certified Kubernetes on your hardware, with Private Nodes as the production default for dedicated GPU nodes.
Yes. Hosted control planes run tenant control planes on a management cluster, which settles where a control plane lives. The management cluster is itself an OpenShift cluster and each hosted cluster is a full OpenShift cluster, so the platform and its subscription follow every tenant. vCluster tenant clusters are vanilla CNCF-certified Kubernetes, which is what AI teams arriving from EKS or GKE expect to be handed.
vCluster offers a spectrum. Private Nodes, the production default, give each tenant dedicated physical nodes with their own CNI and CSI, joined privately over an encrypted WireGuard VPN, so no other tenant schedules onto that hardware. Shared nodes serve development, testing, CI/CD, and trusted teams. vNode layers kernel-native workload isolation using seccomp, cgroups, namespaces, and AppArmor over either model.
Yes, with Private Nodes. Each tenant cluster runs its own CNI and storage, so one customer can run SR-IOV with Multus for RDMA over InfiniBand while another runs Calico with eBPF and a third runs KubeOVN. For AI clouds selling high-performance interconnect, that flexibility decides which workloads you can take.
Yes. Every tenant cluster is CNCF-certified with 100% API compatibility and listed in the Kubernetes conformance repository, with conformance tests run for every major and minor release. Tenants get full cluster-admin rights and install their own CRDs, operators, and GPU tooling without affecting anyone else.
Boost Run launched a managed Kubernetes offering in under 45 days with no new platform engineering hires, and Lintasarta launched a GPU cloud in Indonesia in 90 days. vMetal handles bare metal provisioning from PXE boot through OS installation and network automation via Netris, with vCluster Standalone running as a single binary on top, so the path from rack to tenant cluster is one stack.
See how GPU clouds sell vanilla Kubernetes on their own hardware.
