platform-eng
DGX SuperPOD Kubernetes for AI Cloud Providers
vCluster Platform, named in the NVIDIA DGX SuperPOD reference architecture, deploys hundreds of isolated, CNCF-certified tenant clusters on bare metal for maximum GPU utilization.
Trusted by the fastest-growing AI cloud providers
Problem
The DGX SuperPOD Kubernetes Gap
Raw DGX SuperPOD performance means nothing without the right Kubernetes orchestration layer.
Namespace Isolation Is Too Weak
Tenants can see platform internals they should not. Namespace isolation leaves every customer sharing the same blast radius.
Separate Clusters Are Too Expensive
Provisioning full physical clusters per tenant burns GPU capacity and delays onboarding by weeks, not hours.
Customers Expect a Cloud Experience
AI teams have used AWS and GCP. They expect self-service Kubernetes and cloud-native tooling or they go back to a hyperscaler.
Solution
vCluster: The DGX SuperPOD Kubernetes Layer
vCluster Platform virtualizes the Kubernetes control plane itself, running CNCF-certified tenant clusters as lightweight pods on your DGX SuperPOD bare metal. Every tenant gets their own API server, etcd, and RBAC in seconds, not weeks. For production deployments, Private Nodes, dedicated worker nodes with per-tenant CNI and storage, deliver hardware-level isolation. Named in the NVIDIA DGX SuperPOD reference architecture, production-proven across 100K+ GPU nodes.
Built for DGX SuperPOD Scale
Purpose-built capabilities that turn DGX SuperPOD infrastructure into a production-grade managed Kubernetes service for AI cloud providers.
Tenant Isolation
Isolated Tenant Clusters on Bare Metal
Each tenant gets a fully isolated Kubernetes control plane running as a lightweight pod on your DGX SuperPOD control plane cluster. Own API server, etcd, scheduler and RBAC, spinning up in seconds with near-zero overhead.
- Own API server per tenant
- Spins up in seconds
- Zero physical cluster provisioning
Bare Metal
Zero-Touch DGX Node Provisioning
PXE boot, OS installation, machine registration, and full GPU server lifecycle management from rack to production. Automate the entire DGX SuperPOD provisioning workflow without manual intervention.
- PXE boot and OS install
- Full GPU lifecycle management
- Zero-touch from rack to cluster
Standards
CNCF-Certified Kubernetes Per Tenant
Every DGX SuperPOD tenant cluster is a fully CNCF-conformant Kubernetes control plane with 100% API compatibility. Tenants run any K8s workload without proprietary constraints or compatibility compromises.
- 100% Kubernetes API compatibility
- CNCF conformance certified
- No proprietary API lock-in
AI Platforms
Pre-Validated AI Environments on DGX
Turn a bare DGX SuperPOD Kubernetes cluster into a production AI platform in minutes. Partner integrations with Run:AI, Ray, and Jupyter are certified to work with tenant isolation out of the box.
- Run:AI and Ray integrations
- Cluster to AI platform in minutes
- Certified with tenant isolation
Node Isolation
Dedicated GPU Nodes Per Tenant
Give high-value DGX SuperPOD tenants fully dedicated physical nodes with their own CNI and CSI. No workloads from other tenants share the same hardware, significantly reducing noisy-neighbor GPU contention.
- No cross-tenant hardware sharing
- Own CNI and CSI per tenant
- Eliminates noisy-neighbor contention
Why vCluster
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.
100K+
GPU Nodes Powered
50+
GPU Clouds & F500s
<45
Days to Launch
30K
GitHub Stars
Get Started in 3 Steps
1
Schedule a Demo
Talk to our team about your stack
2
Deploy vCluster
Deploy vCluster on your infra in minutes
3
Onboard Your Tenants
Go live with a hyperscaler-grade tenant experience in days
FAQs
Is vCluster officially part of the NVIDIA DGX SuperPOD reference architecture?
Yes. vCluster Platform is named in the NVIDIA DGX SuperPOD reference architecture.
How does vCluster deploy Kubernetes on DGX SuperPOD bare metal?
vCluster runs CNCF-certified tenant clusters as lightweight processes inside a control plane cluster on your DGX SuperPOD bare metal. Each tenant gets their own Kubernetes API server, etcd, scheduler, and RBAC without requiring separate physical clusters. vCluster Standalone can also run as a single binary directly on bare metal with no external Kubernetes dependency, eliminating the need for k3s or kubeadm as a base layer.
How is tenant isolation enforced on shared DGX SuperPOD infrastructure?
vCluster provides a spectrum of isolation options tailored to DGX SuperPOD deployments. At the control plane level, every tenant has a fully isolated API server and etcd. At the compute level, the production default is private dedicated nodes per tenant; shared nodes are available for dev, test, CI/CD, and trusted-team workloads; and dedicated VMs are also supported. vNode adds workload isolation using seccomp, cgroups, namespaces, and AppArmor, preventing container breakout and limiting blast radius while preserving near-bare-metal performance.
How long does it take to launch a managed Kubernetes service on DGX SuperPOD?
With vCluster Platform and vMetal bare metal provisioning, AI cloud providers have launched production managed Kubernetes services in weeks, not quarters. Boost Run launched their managed Kubernetes offering in less than 45 days. Lintasarta launched a GPU cloud in Indonesia (a GPU cloud in Indonesia) in 90 days, running hundreds of tenant clusters in production from day one.
What AI platforms are supported on DGX SuperPOD Kubernetes with vCluster?
vCluster's Certified Stacks include partner integrations with Run:AI, Ray, and Jupyter that are certified to work with vCluster tenant isolation on DGX SuperPOD infrastructure. Slurm-on-Kubernetes via Slinky integration is also supported for teams running hybrid Slurm and Kubernetes workloads. These integrations turn a bare Kubernetes cluster into a production AI platform in minutes rather than weeks.
Can vCluster handle fleet management across multiple DGX SuperPOD deployments?
Yes. vCluster Platform includes a central UI, CLI, and API for managing all tenant clusters across your entire DGX SuperPOD fleet, including multi-region and multi-data-center deployments. Features include SSO, quotas, templates, auto-sleep, and full Day 2 operations such as observability, updates, backups, disaster recovery, and compliance management, all from a single control plane.
Launch DGX SuperPOD Kubernetes in Weeks
See how AI cloud providers deploy isolated tenant clusters on DGX SuperPOD bare metal.
