Preorder NVIDIA Rubin NVL8 2U Liquid-Cooled AI Server

Architectural Breakthrough: Rubin NVL8

The NVIDIA Rubin NVL8 platform represents a fundamental rethinking of AI server architecture. Instead of treating GPUs, networking, memory, and power as loosely coupled components, NVL8 unifies them into a single, purpose-built compute fabric. The result is an ultra-dense 2U system that delivers hyperscale-class performance in a fraction of the physical footprint.

Rubin NVL8 Unified Compute Fabric

Collapses GPU, networking, memory, and power into a single, purpose-built AI plane—eliminating rack-level bottlenecks and enabling true node-scale supercomputing in 2U.

Native 8-GPU HGX Baseboard

Delivers extreme parallelism with direct GPU-to-GPU pathways, reducing latency and unlocking linear scaling for frontier training and inference workloads.

Integrated 800G Networking Plane

QCT's new air-cooled powerhouse, perfect for demanding data centers.

Direct-to-Chip Liquid Cooling Architecture

Maintains sustained peak performance under full load, removing thermal ceilings that limit air-cooled systems and enabling continuous high-density compute.

54V DC Power Backbone (Up to 24kW)

Purpose-built for hyperscale energy delivery in a single chassis, supporting next-generation GPUs without compromise or derating.

Gen-6 Intel® Xeon Control Plane

Anchors the NVL8 fabric with high-wattage, high-bandwidth CPUs optimized for orchestration, data movement, and pipeline control—ensuring GPUs operate at full utilization without host-side contention.

System Configuration & Platform Specs

The Rubin NVL8 platform is engineered as a complete, production-grade AI node—ready for rack-scale deployment from day one.

At its core is the NVIDIA NVL8 HGX baseboard, purpose-built for extreme parallelism and sustained performance under full load.

Core Platform

8× NVIDIA Rubin GPUs via NVL8 HGX baseboard
Dual Gen-6 Intel® Xeon® CPUs (up to 350W each)
Up to 32× DDR5 DIMMs (6400 MHz)
Fully direct-to-chip liquid cooling
54V DC busbar architecture Supporting up to 24kW

Networking

East–West Fabric: 8× OSFP ports from onboard CX9
InfiniBand XDR / Ethernet 800G
North–South Expansion: 1× PCIe Gen6 x16 FH3/4L slot
BlueField-4 DPU ready

Storage

Data: 8× hot-swap E1.S NVMe SSDs
Boot: 2× PCIe M.2 2280

Physical

Form Factor: 2U rack mount
Dimensions: 448 × 87 × 800 mm

Built For Frontier Workloads

Rubin NVL8 is designed for organizations operating at the edge of what today’s infrastructure can support. This platform exists for teams that cannot afford thermal throttling, interconnect bottlenecks, or unpredictable scaling limits.

Frontier Model Training & Large-Scale Inference Platforms

Train trillion-parameter models with fewer nodes, lower latency, and sustained peak throughput across long-running jobs. Also, powers real-time, high-volume inference with deterministic performance and rack-level efficiency.

Scientific & HPC Computing

Enable dense, high-bandwidth compute for climate modeling, genomics, physics simulations, and national research workloads.

Expert Configuration and Deployment Support

We ensure your AI infrastructure is perfectly matched to your workloads, with comprehensive support from planning through production deployment.

Workload Analysis & Configuration Planning

Our technical team conducts in-depth analysis of your AI models and performance requirements to deliver a perfectly tuned configuration. We ensure optimal balance of compute, memory, and networking for your specific workloads, maximizing performance and ROI from the start.

Technical Validation & Performance Testing

We perform pre-deployment benchmarking with your actual workloads and frameworks. Our engineers verify compatibility, optimize performance, and validate production readiness, ensuring your system meets all performance targets before deployment.

Deployment & Integration Services

From Kubernetes and Slurm integration to network fabric configuration and security hardening, we handle the complete deployment lifecycle. We ensure seamless integration with your existing data center infrastructure and operational workflows.

Dedicated White-Glove Support

Receive direct access to our support engineers for proactive monitoring, firmware updates, and troubleshooting. We act as your dedicated advocate to ensure peak system performance and uptime throughout the entire system lifecycle.

Streamlined Procurement & Logistics

We simplify the entire procurement process with a single point of contact for quoting, configuration, and global logistics. Navigate complex supply chains and delivery timelines with our expert coordination, reducing deployment lead times and operational overhead.

Ongoing Optimization & Scaling Support

Our partnership continues post-deployment with performance monitoring, scaling consultation, and update management. We help you adapt to evolving workload demands and ensure your infrastructure grows with your AI ambitions.

Reserve Your Rubin Allocation

The Rubin NVL8 platform is entering production against fixed global capacity. Once this window closes, subsequent systems move into later manufacturing cycles.

This preorder secures:

A 2U liquid-cooled NVL8 system
8× NVIDIA Rubin GPU allocations
Priority placement in the September 2026 production run

Preorder Your 2U Liquid-Cooled AI Server With 8× NVIDIA Rubin GPU Allocations