MANAGED KUBERNETES

Ship managed Kubernetes as a product.

Give every customer and every internal team a real, conformant Kubernetes cluster, with its own control plane, its own nodes, and its own admin rights, on infrastructure you own and operate.

Schedule a Demo

Read the Docs

Powering managed Kubernetes for AI clouds and AI factories

THE PROBLEM

Managed Kubernetes is easy to demo and brutal to operate.

OPTION A

A namespace is not a cluster

Tenants need their own CRDs, operators, and API versions.

OPTION B

A cluster each does not scale

Every control plane is another thing to patch, monitor, and pay for.

EITHER WAY

Your platform is showing

kubectl get nodes exposes your nodes, your agents, and everyone else’s infrastructure.

You need cluster-grade isolation with namespace-grade economics.

A virtualized control plane per tenant. One fleet and one pool of hardware for you.

Isolation Model

Virtual control planes, real isolation.

Choose the control plane deployment model that fits your infrastructure.

Control Planes as Pods

Deploy vCluster Standalone on a few bare metal CPU machines as your Control Plane Cluster

Faster spin up time
Less resources needed
Manage tenant control planes the Kubernetes-native way

Control Planes on VMs

Connect any VM provisioning tool to vCluster Platform or let vCluster Platform spin up a KubeVirt cluster

OS-level separation between every tenant
Built for strict compliance requirements
Works with KubeVirt or any VM provisioner

Choose the workload isolation model that matches your security and cost requirements.

Private Nodes

Reserve physical nodes, including GPUs, exclusively per tenant

Dedicated nodes with no hardware sharing between tenants
Full GPU memory isolation, no noisy-neighbor interference
Meets the strictest compliance and data residency requirements

Shared Nodes

Run tenant workloads on shared infra with namespace-level guardrails

Maximum tenant density on shared infrastructure
Namespace isolation with network policies and resource quotas
Lower cost per tenant

Harden control plane pods and workloads

Combine with vNode for runtime-level isolation, or pair with gVisor or Kata Containers. Blocks container escapes and kernel exploits at the node level.

Learn more

TENANT VISIBILITY

Your tenants see only their cluster.
Nothing else.

Most Kubernetes platforms expose control plane nodes, platform agents, and other tenants’ infrastructure to every tenant. vCluster virtualizes the control plane entirely, reducing your risk as a platform operator and giving your tenants the clean, isolated experience they’d get from a hyperscaler.

Standard approach

Standard Control Plane Nodes

No isolation: Tenants see platform internals they shouldn’t, unlike any hyperscaler
Security exposure: Control plane components can be compromised via tenant workloads
Stability risk: Tenants can accidentally delete or modify platform components, causing downtime

With vCluster

Virtualized Control Plane

Runs as

Clean experience: Tenants see only their own cluster: the isolated view they’d expect from a hyperscaler
No security exposure: Control plane and platform components are fully hidden from tenant workloads
No stability risk: Platform infrastructure is protected from accidental changes or deletions by tenants

TENANT EXPERIENCE

A cluster that behaves like a cluster.

No shims, no restricted API subset, no “you can’t install that here.”

Conformant Kubernetes

kubectl, Helm, Argo CD, and every operator, unmodified.

Their own control plane

Dedicated api-server, scheduler, and datastore per tenant.

Admin rights, safely

Install CRDs, register webhooks, hold cluster-admin.

Version choice per tenant

Different Kubernetes versions side by side.

Private nodes, including GPUs

Dedicated bare metal or GPU nodes. No noisy neighbors.

Their networking and storage

On private nodes, their own CNI and CSI.

OPERATOR EXPERIENCE

Fleet operations that stay flat as you add tenants.

Your hundredth tenant should cost what your tenth did.

Provision in seconds

Seconds, not the tens of minutes a full cluster takes.

One console for the fleet

Every tenant cluster in one console. RBAC, SSO, and audit built in.

Density that tracks real usage

Pack many control planes onto one cluster. Sleep the idle ones.

Upgrades you control

Roll versions per tenant, per cohort, or fleet-wide.

Quotas, cost, and chargeback

Per-tenant limits and usage data, GPU hours included.

API and GitOps first

Every cluster is a Kubernetes resource. Terraform, Argo CD, your portal.

FLEET MANAGEMENT

Your fleet is a Git repo, not a runbook.

vCluster Platform registers every tenant cluster with Argo CD or Akuity automatically.

Clusters register themselves

Reference a connector by name. Delete the cluster and Platform deregisters it.

Applications ship with the cluster

ArgoCDApplication and ArgoCDApplicationTemplate live in the tenant’s own vcluster.yaml. Workloads land on day 0.

One template, the whole fleet

Define once, reference everywhere. A version bump reaches every cluster on the next sync.

Private clusters included

A tenant’s API server never has to be reachable from Argo CD.

Read: Running production tenant fleets with vCluster and Argo CD

# vcluster.yaml
# the cluster and what runs in it, together
argoCD:
  applications:
    name: monitoring
    target: vcluster
    template:
      name: base-observability
    name: gpu-operator
    target: vcluster
    template:
      name: nvidia-gpu-operator

# app-template.yaml
# defined once, referenced by every tenant
kind: ArgoCDApplicationTemplate
metadata:
  name: base-observability
spec:
  template:
    spec:
      source:
        repoURL: "https://github.com/acme/charts"
        targetRevision: "main"
        path: observability
      syncPolicy:
        automated:
          prune: true
          selfHeal: true

Bump targetRevision once. Every tenant picks it up on the next sync.

DAY 2 AND BEYOND

Nothing breaks on day one.
It breaks on day two hundred.

Launch is the easy part. What decides whether a managed Kubernetes offering survives is the load that accumulates behind it.

Every tenant needs a control plane, nodes, and a network.

Virtualized control planes in seconds, private nodes attached per tenant.

Nodes fail, tenants resize, someone needs a restore.

Add, drain, and retire nodes per tenant. Snapshot and restore one cluster.

Tenants sit on different versions and nobody agrees on a window.

Roll versions per tenant or per cohort. HA control planes self-heal.

Config drift, no per-tenant audit trail, unclear cost per customer.

One vcluster.yaml in Git. Per-tenant metrics, audit, and usage.

“Do we spend multiple millions getting a Kubernetes stack up and running on our own, or do we partner with a firm that has actually done this before?”

Harry Georgakopoulos
COO, Boost Run

PROOF

The best in the industry trust vCluster.

NVIDIA Preferred Cloud Service Provider delivering enterprise-grade GPU infrastructure across multiple U.S. data centers.

SOC 2

ISO 27001

ISO 27701

HIPAA

View case study

<45

Days from decision
to production launch

New platform engineering
hires required

Boost Run took a GPU-native managed Kubernetes service to market on bare metal, using the Private Nodes model for dedicated GPU isolation per tenant and Netris for automated tenant-level network isolation. Tenants can run GPU Operator, Ray, KServe, AI schedulers, and their own CRDs inside their isolated cluster.

Private Nodes

Netris network isolation

GPU Operator

Ray

KServe

Custom CRDs

Speed to market is everything in the GPU cloud space. Partnering with vCluster was a deliberate strategic decision.

Andrew Karos
CEO, Boost Run

Guide

Launch a Managed Kubernetes Platform on Bare Metal in 2 Weeks

Learn how to ship a production-ready managed Kubernetes platform with strong isolation, GPU scheduling, and enterprise-grade operations from day one.

View guide

SOLUTION

Automate Network Isolation for Hard Multi-Tenant Kubernetes

vCluster and Netris integrate Kubernetes and network automation.

View solution

GUIDE

vCluster Guide to Achieve ClusterMAX™ Platinum Rating

Learn how to deliver enterprise-grade Kubernetes for AI workloads and improve ClusterMAX™ rating.

View guide

THE SAME MACHINERY

Managed Kubernetes is the first product, not the only one.

Same control plane virtualization, same isolation model, same fleet tooling. Same vcluster.yaml, same Argo CD templates, same Platform console. The next cluster type is a template, not a platform rebuild.

Slurm Clusters

Tenants see only their own partition and jobs.

BETA

Run:AI Clusters

Tenants see only their own projects and jobs.

AVAILABLE

Ray Clusters

Tenants see only their own head node and workers.

AVAILABLE

Inference Clusters

Dynamo, llm-d, and more.

SOON

One API no matter what the hardware underneath demands

Add new server vendors, NIC types, or GPU generations without changing how clusters get built. The provisioning layer is future-proofed too.

Explore vMetal

Launch your managed Kubernetes platform.

Talk to a Kubernetes expert and get a live walkthrough built around your infrastructure and your tenants.