Introducing vMetal: Run Your GPU Data Center Like a Hyperscaler

The race to build AI infrastructure is accelerating.

Across the industry, organizations are deploying massive GPU clusters to power the next generation of AI applications. New Neocloud providers are emerging, enterprises are building internal AI factories, and demand for GPU infrastructure continues to surge.

But while buying GPUs has become easier, operating them like a cloud platform is still incredibly difficult.

Selling raw GPU infrastructure is quickly becoming a commodity. To stand out and maximize GPU utilization, providers must deliver something more: a managed platform experience similar to EC2 or EKS, where teams can spin up environments and start running workloads immediately.

Building that experience requires a complex stack of infrastructure systems, from machine provisioning to cluster orchestration to tenant environments and AI platforms. Many of the end-to-end platforms designed to manage infrastructure date back nearly two decades, while newer open source tools tend to solve only individual parts of the problem. As organizations rapidly build new GPU data centers and AI factories, the pace of infrastructure deployment has outgrown the tooling available today. As a result, most organizations end up attempting to stitch together a mix of legacy platforms, open source tools, and custom automation.

At vCluster Labs, we believe AI infrastructure should operate as a unified platform, not a collection of disconnected tools. Today, we’re introducing vMetal, a new bare metal provisioning and lifecycle management layer designed to help Neocloud providers and AI factories turn raw GPU hardware into programmable infrastructure.

The Problem: Operating GPU Infrastructure Is Hard

Buying GPUs is only the first step. Operating them like a cloud platform is another story.

Organizations building GPU infrastructure are expected to deliver an experience similar to hyperscalers, where teams can spin up environments on demand and run workloads immediately. But achieving that experience requires infrastructure capabilities across several layers:

• Bare metal provisioning and hardware lifecycle management
• Network orchestration across clusters and tenants
• Kubernetes cluster operations
• Tenant isolation and environment provisioning
• AI tooling and GPU scheduling platforms

Most organizations attempt to build these capabilities internally or combine multiple tools to approximate them. But building a GPU cloud platform from scratch takes significant engineering effort and time. And time matters. A $10M GPU cluster generating several dollars per GPU hour can lose millions in potential revenue if platform launch is delayed by months.

The challenge is not hardware. It is infrastructure automation. We built vMetal to solve that problem.

Introducing vMetal: Bare Metal Provisioning for AI Infrastructure

vMetal is a new machine management layer within the vCluster Platform that automates the lifecycle of bare metal GPU servers.

It transforms physical infrastructure into programmable capacity that can be provisioned, assigned, upgraded, and repurposed through a centralized control plane. Instead of manually configuring machines or building custom provisioning pipelines, infrastructure operators can manage their entire GPU fleet through a unified system.

With vMetal, organizations can:

• Automatically discover machines connected to the network
• Provision servers via PXE boot
• Manage machine lifecycle events such as upgrades or reconfiguration
• Assign machines directly to Kubernetes clusters or infrastructure pools
• Prepare machines for multi-tenant environments

The result is a system where bare metal behaves more like cloud infrastructure. Servers become resources that can be allocated, reassigned, and managed through software workflows rather than manual operations.

From Rack to Cluster in Minutes

Bringing new GPU hardware online is traditionally slow and manual. Servers often require installation, configuration, and networking setup before they can even be attached to a cluster.

vMetal automates this entire process. Using automated provisioning and PXE-based bootstrapping, machines can move from rack installation to cluster-ready nodes in minutes.

Infrastructure operators can:

• Power on machines
• Automatically install operating systems
• Apply configuration and networking policies
• Attach nodes to Kubernetes clusters

All through the vCluster Platform.

This dramatically reduces the time required to expand GPU capacity and allows infrastructure teams to operate clusters with the speed and flexibility expected from modern cloud environments.

Certified Stacks: Pre-Validated AI Platforms You Can Deploy in One Command

Provisioning infrastructure is only part of the challenge. Platform teams still need to assemble the tooling required for real AI workloads, including GPU scheduling systems, platform services, and AI development frameworks. That is where vCluster Certified Stacks come in.

Certified Stacks provide tested and maintained blueprints for deploying AI-ready platforms, combining:

• vCluster tenancy configurations
• Kubernetes platform components
• GPU scheduling and workload orchestration
• AI tooling and development environments

These stacks allow platform teams to deploy complete AI environments quickly while still retaining the flexibility to customize their infrastructure.

The first Certified Stacks support a growing ecosystem of AI infrastructure platforms, including:

• NVIDIA Run:ai for enterprise GPU orchestration
• SkyPilot for running and scaling AI workloads across infrastructure
• Ray for distributed AI applications and model training
• Slinky for AI platform orchestration

Each stack is delivered as a maintained Terraform blueprint, enabling teams to go from infrastructure to a working AI platform in a repeatable and reliable way.

The New Infrastructure Stack for AI

With the introduction of vMetal and vCluster Certified Stacks, the vCluster ecosystem now spans the full infrastructure stack required to run AI workloads. Organizations building GPU clouds or enterprise AI platforms can deploy a layered architecture designed specifically for AI infrastructure.

• vMetal: Bare metal machine management that provisions and operates GPU servers.
• vCluster: Tenant and cluster orchestration that enables multiple teams or customers to safely share Kubernetes infrastructure.
• vNode: Secure runtime isolation for AI workloads running inside shared clusters.
• vCluster Certified Stacks: Preconfigured AI environments that combine GPU scheduling, platform services, and AI tooling.

Together, these layers create a unified system capable of running AI workloads from physical machines all the way up to production AI environments.

Turn Your GPU Racks Into a Cloud Platform

Running AI at scale requires more than powerful hardware. It requires infrastructure capable of coordinating machines, clusters, tenants, and workloads across an entire platform.

With the introduction of vMetal and Certified Stacks, the vCluster ecosystem now provides a unified stack for AI infrastructure: Bare Metal → Kubernetes → Tenant Environments → AI Platforms

Instead of stitching together dozens of tools, platform teams can now build AI infrastructure using components designed to work together from the start.

If you’re building GPU infrastructure for a Neocloud or AI factory and want to learn more, visit vMetal.