🚀 ANNOUNCEMENT: CambridgeNexus Partners with ProphetStor to Deliver 1.5x Performance Gains on CNEX AI Factory •

🚀 ANNOUNCEMENT: CambridgeNexus Partners with ProphetStor to Deliver 1.5x Performance Gains on CNEX AI Factory •

🚀 ANNOUNCEMENT: CambridgeNexus Partners with ProphetStor to Deliver 1.5x Performance Gains on CNEX AI Factory •

NVIDIA Certified System · Blackwell Ultra

The Next Era of AI Compute: GB300 NVL72

The CNEX AI Factory is a fully managed, rack-scale compute environment built on the NVIDIA GB300 NVL72. Bare-metal GPU capacity, high-speed AI fabric, enterprise storage, and 24×7 operations — purpose-built for frontier AI.

RESERVE GB300 CAPACITY

NVIDIA GB300 NVL72

Enterprise LOI submitted

~150 kW / RACK

Enterprise LOI submitted

NVL72 COHERENT FABRIC

Enterprise LOI submitted

The Next Era of AI Compute:

GB300 NVL72.

CNEX operates dedicated AI Factory infrastructure

built on the NVIDIA NVL72

The Next Era of AI Compute: GB300 NVL72. CNEX operates dedicated AI Factory infrastructure built on the NVIDIA NVL72 rack-scale system.

Up to 99.95%

Availability SLA Guaranteed

Up to 99.95%

Availability SLA Guaranteed

1

GPUs

NVIDIA GB300 NVL72

Full NVL72 single coherent fabric with 13.5 TB HBM3e Memory.

1

GPUs

NVIDIA GB300 NVL72

Full NVL72 single coherent fabric with 13.5 TB HBM3e Memory.

1Gb/s

InfiniBand per GPU

1Gb/s

InfiniBand per GPU

Purpose-built for frontier AI model training, fine-tuning, large-scale inference, and sovereign AI deployments.

RESERVE GB300 CAPACITY

AI Fabric

800 Gb/s per GPU via NVIDIA Quantum-X800 InfiniBand with Ultra-low-latency.

AI Fabric

800 Gb/s per GPU via NVIDIA Quantum-X800 InfiniBand with Ultra-low-latency.

Compute Power

36 NVIDIA Grace CPUs delivering 2,592 cores and 17+ TB Unified Memory.

Compute Power

36 NVIDIA Grace CPUs delivering 2,592 cores and 17+ TB Unified Memory.

Power & Cooling

~145–150 kW per NVL72 rack managed by advanced Direct Liquid Cooling (DLC).

Power & Cooling

~145–150 kW per NVL72 rack managed by advanced Direct Liquid Cooling (DLC).

Multi-Site Scale

Supported multi-rack scalability and federation globally via AboveCloud™.

Multi-Site Scale

Supported multi-rack scalability and federation globally via AboveCloud™.

Key Platform Metrics

Frontier-scale capacity, in a single rack

Every CNEX AI Factory deployment delivers a fully integrated NVL72 system — compute, memory, fabric, and cooling validated as one.

13.5 TB

HBM3e GPU Memory

17+ TB unified GPU + CPU coherent memory.

13.5 TB

HBM3e GPU Memory

17+ TB unified GPU + CPU coherent memory.

13.5 TB

HBM3e GPU Memory

17+ TB unified GPU + CPU coherent memory.

800 Gb/s

InfiniBand per GPU

NVIDIA Quantum-X800, 4-rail AI fabric.

800 Gb/s

InfiniBand per GPU

NVIDIA Quantum-X800, 4-rail AI fabric.

800 Gb/s

InfiniBand per GPU

NVIDIA Quantum-X800, 4-rail AI fabric.

DLC

Primary cooling at ~145–150 kW per rack.

DLC

Primary cooling at ~145–150 kW per rack.

DLC

Primary cooling at ~145–150 kW per rack.

2,592

Grace CPU Cores

36 NVIDIA Grace CPUs · Arm Neoverse V2.

2,592

Grace CPU Cores

36 NVIDIA Grace CPUs · Arm Neoverse V2.

99.95%

Availability SLA

Fully managed 24×7 NOC operations.

99.95%

Availability SLA

Fully managed 24×7 NOC operations.

Compute & Networking

Engineered as one coherent system

All 72 GPUs operate within a single NVLink domain, interconnected by a 4-rail Quantum-X800 InfiniBand fabric for near-zero-latency GPU-to-GPU communication.

Compute Architecture

Blackwell Ultra + Grace

GPU Architecture

NVIDIA Blackwell Ultra

GPU Count

72 × NVIDIA GB300

NVLink Domain

Full NVL72 coherent fabric

GPU Memory

13.5 TB HBM3e

CPU Architecture

NVIDIA Grace · Arm Neoverse V2

CPU Count

36 Grace CPUs · 2,592 cores

Unified Memory

17+ TB (GPU + CPU, coherent)

Operating System

Ubuntu 24.04 LTS

Networking & AI Fabric

Quantum-X800 InfiniBand

AI Fabric

NVIDIA Quantum-X800 InfiniBand

Fabric Bandwidth

800 Gb/s per GPU

Fabric Configuration

4-Rail InfiniBand

East-West Latency

Ultra-low-latency GPU-to-GPU

Data Processing Unit

NVIDIA BlueField-3

North-South Uplink

200G / 400G Ethernet

Multi-Rack Scalability

Supported

Multi-Site Federation

AboveCloud™

AI Fabric

800 Gb/s per GPU across a 4-rail InfiniBand topology, scalable across racks and federated across sites.

AI Fabric

800 Gb/s per GPU across a 4-rail InfiniBand topology, scalable across racks and federated across sites.

AI Fabric

800 Gb/s per GPU across a 4-rail InfiniBand topology, scalable across racks and federated across sites.

Supported Workloads

Built for every frontier workload

From multi-trillion-parameter training to latency-sensitive inference and sovereign deployments — matched to the segments that run them.

Frontier Model Training

Multi-trillion-parameter foundation model runs

AI Labs · Research Institutions

Enterprise Fine-Tuning

Domain-adapted models for production deployment

Enterprise AI Teams

Production Inference

High-throughput, latency-sensitive serving

SaaS · API Providers

Agentic AI

Multi-agent orchestration and reasoning systems

Enterprise Software

Digital Twins

Industrial and geospatial simulation at scale

Manufacturing · Energy

Physical AI / Robotics

Autonomous systems training and simulation

Robotics · Automotive

Sovereign AI

Dedicated national or enterprise AI environments

Government · Regulated Industries

HPC / Scientific Computing

AI-HPC convergence workloads

Life Sciences · National Labs

Reserve your GB300 NVL72 capacity

Deploy as Dedicated Bare Metal, a Private AI Factory, or a Sovereign AI Cloud — fully managed or customer-managed, in Tier III+ facilities.

RESERVE GB300 CAPACITY

CORTEX™

AI Factory Orchestration

ABOVECLOUD™

Multi-Site Mobility

24×7 NOC

Managed Cluster Operations

REST · Terraform

API & Automation Layer

The Next Era of AI Compute: GB300 NVL72

The Next Era of AI Compute: GB300 NVL72