PREVIEW: How AI Cloud Provider Corvex Delivers Dedicated, Isolated Kubernetes Clusters to Enterprise AI Customers with vCluster

Building Secure AI Cloud Infrastructure Demands More Than Bare Metal

Corvex is a next-generation AI cloud provider, formerly known as Klustr.ai, delivering high-performance GPU infrastructure for enterprise AI workloads. With a focus on secure, high-performance inference and deep expertise in the Kubernetes ecosystem, Corvex has established itself as a trusted partner for organizations building at the frontier of AI. Its team knows the space, knows the players, and came to market with a clear vision: go beyond raw compute and deliver a complete platform experience.

As enterprise AI adoption accelerated, Corvex's customers made one thing clear: bare metal GPU access alone was no longer enough. They expected fully managed, dedicated Kubernetes environments with strong isolation, self-service provisioning, and native GPU support. Expanding into managed Kubernetes was the obvious next step for Corvex. The challenge was how to architect it correctly on bare metal GPU infrastructure.

The Architecture Challenge: A Dedicated Cluster Per Customer on Bare Metal GPUs

When a strategic enterprise customer came to Corvex requiring Kubernetes on its GPU infrastructure, the requirements were specific and demanding. The customer needed:

• A fully isolated Kubernetes cluster of their own with complete workload separation and no interference from other customers on the same infrastructure
• Dedicated GPU nodes with predictable performance for AI training and inference and no noisy-neighbor risk
• Self-service administration allowing the customer's team to create, modify, and scale workloads independently, without requiring the Corvex team to be available around the clock
• Automated node provisioning on OpenStack infrastructure, without manual intervention for each deployment
• Controlled permissions boundaries that gave the customer's team meaningful control without exposing the underlying control plane cluster

Building a solution to satisfy all of these requirements from scratch, covering per-customer cluster isolation, dedicated GPU node attachment, automated provisioning, granular RBAC, and network segmentation, would have required months of platform engineering. Corvex needed a production-ready foundation, not a development project.

One Dedicated Cluster Per Customer, Provisioned on Bare Metal

vCluster offered exactly the architecture Corvex needed: each customer gets their own fully isolated Kubernetes control plane, backed by dedicated GPU nodes, provisioned automatically on Corvex's OpenStack bare metal infrastructure. Corvex deployed vCluster as the foundation of its managed Kubernetes offering and moved from evaluation to production with its first enterprise customer.

• Private Nodes: Dedicated GPU Hardware Per Customer Cluster Through vCluster's Private Node model, each of Corvex's customers receives a Kubernetes cluster backed by GPU hardware dedicated exclusively to them. Each customer cluster is fully isolated with dedicated resources, no resource contention, and no operational bleed between customers. Private Nodes were the single capability that made the first enterprise deployment possible.
• Auto Nodes: Automated Cluster Provisioning on OpenStack vCluster's Auto Nodes capability enabled Corvex to automate GPU node provisioning directly on its OpenStack bare metal infrastructure. Nodes are provisioned dynamically as customer workload demands grow, eliminating the manual overhead that would otherwise bottleneck onboarding and scaling. Auto Nodes and Private Nodes together were the specific combination that unlocked Corvex's first enterprise deployment.
• Netris Integration: Automated Network Isolation Per Cluster vCluster's integration with Netris provides automated, per-cluster network provisioning. Each customer environment receives its own isolated network paths without requiring manual configuration, enabling Corvex to onboard new customers rapidly while maintaining the security boundaries enterprise AI workloads require.
• Granular RBAC: Customer Self-Service Without Keys to the Kingdom vCluster's flexible RBAC model gives Corvex's customers meaningful self-service control, including the ability to create and manage workloads, configure resources, and operate their environments independently, without exposing the underlying control plane cluster. Custom RBAC objects applied through vCluster templates enable fine-grained permission boundaries, including persistent volume claim management at the namespace level. Corvex is also deploying SSO to further streamline customer access management.

The result is a managed Kubernetes platform where every customer gets their own cluster, their own dedicated GPU hardware, and their own isolated network, all provisioned automatically on Corvex's bare metal infrastructure.

The Right Architecture at the Right Moment

Corvex had been tracking vCluster's development since its earliest days as Klustr.ai, building a relationship over eight months before the deal closed. When a strategic enterprise customer arrived with specific Kubernetes requirements on bare metal GPU infrastructure, the timing was right. Private Nodes and Auto Nodes, newly launched, were exactly the capabilities needed to make the deployment work.

• The right features at the right time. The simultaneous availability of Private Nodes for dedicated GPU isolation and Auto Nodes for automated provisioning on OpenStack created the architecture Corvex needed. Neither capability alone would have been sufficient; together, they made the enterprise deployment possible.
  
• A production-ready foundation without the build cost. Delivering per-customer dedicated Kubernetes clusters on bare metal from scratch, with isolated control planes, dedicated GPU node attachment, automated provisioning, and granular RBAC, would have taken months and significant engineering investment. vCluster provided a production-hardened foundation Corvex could deploy against a live customer opportunity.
  
• Better value than the alternatives. vCluster delivered more capability at a better price point than the alternatives Corvex evaluated. In a market where GPU infrastructure economics are central to the business model, the cost efficiency of the vCluster architecture made the partnership commercially compelling.
  
• Real isolation, without the operational overhead. Corvex's customers get their own cluster, their own nodes, and their own network. vCluster makes that possible without Corvex having to manage a separate physical cluster for every customer or take on the operational complexity that would normally come with it.
• ‍A proven AI cloud architecture. The vCluster + Netris + OpenStack stack that Corvex deployed has become a reference architecture for GPU-native AI cloud providers, with a growing ecosystem of AI cloud providers now building their managed Kubernetes platforms on vCluster.

Looking Ahead

With managed Kubernetes live in production, including a 32-node dedicated cluster serving its first enterprise customer, Corvex has established itself as a full-stack AI cloud provider. The platform is now the foundation for bringing on additional enterprise customers, each receiving their own dedicated cluster backed by Corvex's bare metal GPU infrastructure.

Corvex is building toward GitOps-driven cluster management, expanding its observability stack, and exploring advanced storage architectures to support increasingly complex enterprise AI workloads. The vCluster partnership provides the architectural foundation to scale without rebuilding.

For AI cloud providers evaluating how to deliver managed Kubernetes on bare metal GPU infrastructure, the Corvex and vCluster partnership offers a clear reference: give every customer their own cluster, back it with dedicated GPU hardware, and automate the provisioning, all without building it from scratch.

"We use OpenStack, launch our VMs, and deploy platform services directly on top. What vCluster gives us is the ability to hand each customer their own Kubernetes environment, fully isolated with their own control plane, without that complexity bleeding back into our infrastructure. We're building toward a platform that's repeatable and reliable at scale, and vCluster is the foundation of that." -Morgan Sanford, Principal DevOps Engineer, Corvex