ai-cloud
Bare Metal Karpenter GPU Autoscaling for AI Clouds
vMetal Auto Nodes brings Karpenter-style bare metal GPU provisioning to your AI infrastructure. Scale GPU server fleets on demand without VM overhead or hypervisor tax.
Trusted by the fastest-growing AI cloud providers
Problem
GPU Scaling Without Bare Metal Is Broken
Standard Kubernetes autoscaling wasn't built for bare metal GPU server fleets.
VM Overhead Kills GPU Performance
Hypervisors and virtual machines consume GPU cycles, memory, and latency your AI workloads cannot afford to lose.
Karpenter Doesn't Provision Bare Metal
Karpenter was designed for cloud VMs. Extending it to physical GPU servers requires custom engineering your team hasn't started.
Manual Provisioning Slows Your Revenue
Every hour a GPU rack sits unscheduled while teams manually provision servers is revenue your AI cloud is not generating.
Solution
Auto Nodes: Bare Metal Karpenter for GPU Servers
vMetal's Auto Nodes brings Karpenter-style on-demand provisioning to physical GPU servers. When tenants schedule workloads, Auto Nodes automatically provisions bare metal GPU nodes via Terraform — no VMs, no hypervisor, no manual intervention. The same lifecycle management covers the full path from PXE boot to decommission.
Full Stack Bare Metal GPU Infrastructure
From physical rack to isolated tenant Kubernetes clusters, every layer runs without VM overhead or manual provisioning steps.
Bare Metal Karpenter
On-Demand GPU Node Provisioning
Auto Nodes automatically provisions bare metal GPU servers via Terraform when tenants schedule workloads. Karpenter-style scaling for physical hardware — no VMs, no cloud dependency, no manual steps.
- Terraform-driven bare metal GPU provisioning
- Scales in response to workload demand
- Zero manual intervention required
Bare Metal Lifecycle
Zero-Touch GPU Server Provisioning
vMetal handles PXE boot, OS installation, machine registration, and network configuration end to end. GPU racks go from physical hardware to production-ready Kubernetes nodes without operator involvement.
- PXE boot and OS install automated
- Full machine lifecycle to decommission
- Network automation via Netris integration
K8s Distribution
Kubernetes Directly on Bare Metal
vCluster Standalone runs as a single binary on bare metal Linux — no k3s, no kubeadm, no external Kubernetes dependency as a base layer. The lightest path from GPU hardware to certified Kubernetes.
- Single binary on bare metal Linux
- Replaces k3s and kubeadm entirely
- CNCF-certified Kubernetes distribution
Workload Isolation
Kernel-Native Isolation at Bare Metal Speed
vNode provides container breakout protection using seccomp, cgroups, namespaces, and AppArmor per workload — with no hypervisor tax. Bare metal GPU performance is fully preserved across isolated tenant environments.
- No hypervisor overhead on GPU performance
- Container breakout protection built in
- seccomp, cgroups, AppArmor per workload
Tenant Isolation
Dedicated GPU Nodes Per Tenant
Each tenant gets fully dedicated physical GPU nodes with their own CNI and CSI. No workloads from other tenants share the hardware — complete node-level isolation for high-stakes AI workloads.
- Fully dedicated physical GPU nodes
- Own CNI and CSI per tenant
- No noisy-neighbor GPU 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
What is karpenter bare metal GPU provisioning and how does it work?
Karpenter bare metal GPU provisioning refers to automatically scaling physical GPU servers in response to workload demand — the way Karpenter scales cloud VMs, but applied to real hardware. vMetal's Auto Nodes feature does exactly this: when a tenant schedules a GPU workload, Auto Nodes triggers Terraform to provision the next available bare metal GPU server, registers it into the cluster, and makes it schedulable — all without manual operator steps or VM intermediaries.
Does vMetal support zero-touch bare metal GPU provisioning from rack to Kubernetes?
Yes. vMetal handles the full lifecycle from PXE boot and OS installation through machine registration, network automation, and Kubernetes node enrollment. GPU servers move from physical rack to production-ready nodes without manual intervention. vCluster Standalone then runs a CNCF-certified Kubernetes distribution directly on the bare metal as a single binary — no k3s or kubeadm required as a base layer.
How does Auto Nodes differ from standard Karpenter for GPU infrastructure?
Standard Karpenter is designed for cloud VMs and interacts with cloud provider APIs to spin up virtual instances. Auto Nodes is purpose-built for bare metal GPU servers — it provisions physical hardware via Terraform when workloads require it. This eliminates the hypervisor layer entirely, preserving full GPU performance for AI workloads while still delivering the automated, demand-driven scaling behavior that Karpenter is known for in cloud environments.
Can multiple tenants share bare metal GPU nodes with strong isolation?
Yes. vCluster supports a flexible isolation spectrum. At the hardware level, Private Nodes give each tenant fully dedicated physical GPU nodes with their own CNI and CSI. At the workload level, vNode adds kernel-native container breakout protection using seccomp, cgroups, namespaces, and AppArmor — without any hypervisor overhead. AI cloud providers can offer different isolation tiers to different customers on the same physical infrastructure.
Is vCluster used in production GPU cloud environments at scale?
Yes. vCluster powers 100K+ GPU nodes in production across 50+ GPU clouds and Fortune 500 customers including CoreWeave and Nscale. vMetal is referenced in the NVIDIA DGX SuperPOD architecture. Lintasarta launched Indonesia's leading GPU cloud in 90 days using the stack. Boost Run launched a managed Kubernetes offering in under 45 days with zero new platform engineering hires.
Does vMetal work with existing network infrastructure for bare metal GPU deployments?
Yes. vMetal includes network automation for bare metal GPU environments via a Netris integration. This covers VLANs, VXLANs, VRFs, ACLs, and DPU policies — providing hardware-enforced network boundaries per tenant. For distributed GPU deployments spanning multiple racks or data centers, vCluster's built-in VPN provides secure connectivity between control planes and worker nodes.
Scale Bare Metal GPUs Without VM Overhead
See how Auto Nodes provisions GPU servers on demand for your AI workloads.
