SuperCloud Director: Operationalizing NeoCloud Infrastructure with NVIDIA NCX Infra Controller

Powered by NVIDIA NCX Infra Controller and NVIDIA Bluefield DPUs

Introduction

Artificial Intelligence infrastructure is entering a new phase of scale.

AI infrastructure is rapidly evolving from experimental clusters into industrialized AI platforms. Enterprises are building AI factories, while a new generation of NeoCloud providers are emerging to commercialize GPU infrastructure as a service.

However, the transition from GPU clusters to revenue-generating AI cloud platforms requires more than hardware acceleration. It requires a programmable infrastructure control layer capable of securely provisioning bare metal, orchestrating tenants, enforcing policies, and automating lifecycle operations across GPU infrastructure.

NeoClouds service several tenant organizations to support workloads such as:

• Foundation model training
• Enterprise AI inferencing services and internal copilots
• Digital twins and simulation environments
• AI-RAN workloads within telecom networks
• Edge AI inference close to users and devices

Supporting these workloads requires more than GPU clusters. It requires a programmable infrastructure layer capable of orchestrating Bare-Metal compute, networking, and storage resources across sites and with zero trust security between Tenants.

At Supermicro, we see GPU NeoClouds as the natural evolution of Cloud Service Providers - The next generation of Clouds, anchored by the need for speed, accelerated compute, and fast time to online based on geo distributed power availability.

Supermicro’s SuperCloud Suite, anchored by SuperCloud Director, provides this control plane layer for NeoCloud operators. When integrated with NVIDIA NCX Infra Controller (formerly Bare Metal Manager), the platform enables providers to stand up secure multi-tenant GPU clouds while maintaining full bare-metal GPU performance.

Together, these technologies provide the operational backbone required for NeoCloud providers and NVIDIA Cloud Partners (NCPs) to deploy scalable GPU-as-a-Service infrastructure.

By integrating tightly with NCX Infra Controller and leveraging NVIDIA BlueField DPUs for zero-trust isolation, SuperCloud Suite, together with Supermicro Datacenter Building Block Solutions (DCBBS) enables NeoCloud operators to deploy, secure, and monetize accelerated infrastructure at scale, anywhere with power availability.

The NeoCloud Infrastructure Stack

Operating a NeoCloud platform requires more than GPU clusters. It requires a layered infrastructure architecture capable of securely provisioning accelerated infrastructure while enabling multi-tenant commercial operations.

The NeoCloud infrastructure model can be understood as five architectural layers:

Each layer plays a distinct role in transforming accelerated infrastructure into programmable AI cloud platforms.

Infrastructure Controller: Automating Accelerated Infrastructure

At the infrastructure layer, NVIDIA NCX Infra Controller provides lifecycle automation for GPU-based bare-metal environments.

This infrastructure controller enables operators to:

• Discover and provision GPU servers
• Automate cluster deployment
• Manage bare-metal infrastructure lifecycle
• Integrate with DPU-based security models
• Deploy infrastructure using infrastructure-as-code workflows

By operating directly at the bare-metal layer, NCX Infra Controller allows NeoCloud providers to maintain maximum GPU performance while still automating infrastructure operations.

NeoCloud Control Plane: Operating the GPU Cloud

Above the infrastructure layer sits the NeoCloud Control Plane, which provides the operational and commercial platform required to deliver GPU infrastructure as a service.

SuperCloud Director operates in this layer.

SuperCloud Director provides the capabilities needed to run a NeoCloud platform, including:

• Tenant onboarding and lifecycle management
• Multi-tenant GPU infrastructure allocation
• Network segmentation and security policy enforcement
• SLA management
• Usage metering and billing integration
• Enterprise tenant onboarding

This layer transforms GPU infrastructure into a commercial AI cloud platform capable of serving multiple enterprise tenants simultaneously.

Enabling Multi-Tenant NeoCloud Platforms

When the NCX Infra Controller and NeoCloud Control Plane operate together, NeoCloud providers can deploy secure multi-tenant GPU infrastructure platforms.

Multi-Tenancy Boundaries for NeoClouds

This architecture enables providers to deliver:

• GPU-as-a-Service platforms
• Sovereign AI infrastructure
• Dedicated AI training clusters
• Enterprise AI development environments
• High-performance inference platforms

Unlike legacy virtualization-based infrastructure, this approach preserves full GPU performance exposure while maintaining strict tenant isolation.

GPU Utilization and Workload Orchestration

Efficient NeoCloud platforms must maximize the utilization of expensive GPU resources.

Through integration with the SuperCloud Developer Console (SDX), operators can implement:

• Multi-Instance GPU (MIG) partitioning
• Workload-aware GPU scheduling
• Tenant-aware capacity allocation
• Automated workload scaling

These capabilities allow NeoCloud providers to increase GPU utilization while maintaining performance guarantees for enterprise workloads.

Integrated AI Infrastructure with DCBBS

SuperCloud Suite integrates with Supermicro Datacenter Building Block Solution (DCBBS) infrastructure.

Through SuperCloud Composer, operators can automate the monitoring, management and optimization of:

• GPU servers
• Connectivity Infrastructure
• Storage infrastructure
• In-Rack Power and Thermal Solutions
• In-Row Power and Thermal Solutions
• Site Infrastructure Solutions
• Turn-key Datacenter and NeoCloud Services
• SuperCloud Software Suite

This enables rapid deployment of AI factories that can scale into NeoCloud platforms.

SuperCloud Director + NVIDIA NCX Infra Controller

Secure Infrastructure Automation for NeoClouds

At the foundation of the NeoCloud lies secure, automated infrastructure provisioning.

NVIDIA NCX Infra Controller provides an API-based microservice designed for site-local, zero-trust lifecycle management of bare-metal systems.

NCX Infra Controller operates according to several fundamental principles:

• The machine itself is considered untrusted
• All monitoring and management occur through out-of-band mechanisms
• Machines must become operational without manual intervention
• Network fabrics remain stable even as tenants change

These principles enable automated provisioning of large-scale GPU clusters while maintaining strict security boundaries.

NCX Infra Controller handles core infrastructure tasks such as:

• Hardware discovery and inventory management
• firmware validation and updates
• Hardware testing and burn-in
• Power control and system lifecycle management
• PXE provisioning and operating system deployment
• IP address allocation and DNS services
• secure node wiping and tenant transitions

Importantly, NCX Infra Controller focuses on infrastructure lifecycle management, not workload orchestration.

SuperCloud Director builds upon this trusted infrastructure layer to deliver a programmable control plane that manages:

• Tenant lifecycle
• AI Grid Management Portal
• Tenant Administration Controller
• Multi-tenant GPU infrastructure
• Kubernetes and VM environment integration.
• Programmatic Network Isolation across Ethernet and Infiniband Networks
• Integrated Networking Services to enable multi-tenant GPU Consumption
• Turn-key integration into Storage Ecosystem at Block, File and Object layers
• Billing and governance frameworks

This combination allows infrastructure providers to transform bare-metal GPU clusters into commercial cloud platforms capable of serving multiple customers simultaneously.

NVIDIA BlueField DPUs and Zero-Trust Infrastructure Isolation

Security and isolation are fundamental requirements for multi-tenant AI infrastructure.

Traditional virtualization platforms implement isolation at the hypervisor or Host OS layer. However, this introduces additional overhead and increases the attack surface as once Host OS is compromised, the attacker can spread across the entire Network

The SuperCloud Director architecture instead leverages NVIDIA NCX Infra Controller and BlueField DPUs to implement hardware-rooted zero-trust infrastructure isolation where the Isolation logic runs outside the Host OS layer, within the Bluefield DPU.

In a NCX Infra Controller managed environment, each host system pairs:

• a Supermicro accelerated server
• an NVIDIA BlueField DPU

The DPU runs a collection of agents that communicates securely with the NCX Infra Controller control plane and enforces policy at the infrastructure level.

This architecture enables:

• Line-rate east-west traffic isolation
• Tenant-level network segmentation
• Secure workload separation
• Programmable infrastructure security policies

Conceptually, this approach is similar to micro-VM isolation models such as AWS Firecracker but implemented at the DPU infrastructure layer with Bare-Metal OS instead of the hypervisor layer.

The result is a system that delivers:

• Full bare-metal GPU performance
• Strong tenant isolation
• Reduced attack surface
• Deterministic infrastructure security

This architecture makes it possible to safely operate multi-tenant AI infrastructure across enterprise, startup, and sovereign workloads.

Customer Spotlight: DigiPowerX

NeoCloud infrastructure providers are already using this architecture to commercialize GPU infrastructure.

DigiPowerX, an AI infrastructure company focused on the deployment of Tier-3 modular data centers powered by owned and controlled energy assets, is building scalable GPU infrastructure platforms to support the growing demand for AI compute.

By combining accelerated infrastructure with modern orchestration platforms, DigiPowerX is positioning itself to support enterprise AI workloads, AI service providers, and sovereign AI initiatives.

As noted by Jagan Jeyapal, CTO of DigiPowerX:

“The transition from Power Infrastructure supremacy to GPU infrastructure to AI cloud platform supremacy, requires deep integration between infrastructure automation and NeoCloud orchestration. Platforms like SuperCloud Director combined with NVIDIA AI infrastructure control technologies provide the operational backbone needed to deliver scalable and secure GPU cloud services, together with our owned and controlled energy assets.”

Monetizing your NeoCloud

The emergence of the NeoCloud model creates new business opportunities for infrastructure operators.

Instead of operating isolated, underutilized GPU clusters, organizations can build programmable infrastructure platforms capable of serving multiple AI workloads simultaneously.

Examples include:

• GPU-as-a-Service platforms
• Sovereign AI clouds
• Enterprise AI development environments
• AI inference marketplaces
• AI-RAN infrastructure hosting

SuperCloud Director integrates with billing systems and governance frameworks to allow NeoCloud operators to programmatically monetize GPU capacity across tenants and workloads.

The Future: Distributed AI Infrastructure

AI infrastructure is undergoing a transformation like the early days of cloud computing.

First came Virtual Machines

Then came Clouds

Now comes the NeoCloud

NeoClouds deliver AI training and inference closer to users, and enable new classes of AI applications, while continuing to serve large Tenants running Training Jobs.

The technologies required to power this future must combine:

• Accelerated computing
• Programmable infrastructure
• Secure multi-tenant isolation
• Automated operations

By integrating SuperCloud Suite with NVIDIA NCX Infra Controller and BlueField DPUs, Supermicro SuperCloud Suite provides the operational foundation for building and scaling NeoClouds.