Uvation MarketplaceMarketplace

Lenovo NVIDIA GB300 NVL72 Rack Scale AI Server

Build high-performance AI factories with 72 B300 GPUs, fifth-generation NVIDIA NVLink and up to 800 Gb/s ConnectX-8 networking. Hybrid liquid cooling and enterprise-grade Lenovo infrastructure provide exceptional density, scalability and efficiency.

72 NVIDIA Blackwell Ultra B300
GPUs per rack
36 NVIDIA Grace
Processors per rack
17.28 TB LPDDR5X
Memory per rack

Rack-Scale Compute for
Demanding AI Workloads

An integrated rack architecture brings GPU compute, processor resources, high-speed networking, and dedicated power infrastructure into one system.

High-Density GPU Architecture

Houses 72 NVIDIA Blackwell Ultra B300 GPUs across 18 compute trays.

Integrated Memory Design

Provides 960 GB LPDDR5X memory per tray, totaling 17.28 TB per rack, alongside HBM3e GPU memory.

Built-In Network Infrastructure

Incorporates ConnectX-8 network boards, BlueField-3, and NVLink Switch trays within the rack-scale configuration.

Unified Rack-Scale AI Architecture

Compute trays, NVLink Switch trays, storage, and network components operate within a consolidated 48U rack design.

Compute plane with 18 trays and NVLink switch fabric
18
compute trays
9
NVLink Switch trays
Up to 8
E1.S PCIe 5.0 x4 NVMe SSDs
1.92 TB internal
M.2 PCIe 4.0 NVMe SSD per compute tray

Workloads

Designed for AI environments that need rack-level GPU capacity, substantial memory resources, and integrated compute and networking.

Large-Scale AI Model Training

This liquid-cooled rack integrates 72 Blackwell GPUs, delivering massive parallel computing power to train next-generation, trillion-parameter large-scale AI models.

GPU-Accelerated AI Computing

Accelerates dense workloads with 72 interconnected Blackwell GPUs, delivering uncompromised parallel processing power for the most demanding enterprise AI computing.

AI Infrastructure Deployment

Streamlines exascale infrastructure with a turnkey, liquid-cooled rack, delivering rapid deployment and maximum power efficiency for elite enterprise AI datacenters.

Engineered to dominate

Compute, Memory, and
Power at Rack Scale

A high-density configuration combines Blackwell Ultra GPU accelerators, Grace processors, rack-level memory, and dedicated power shelves.

Exploded view of the NVIDIA GB300 NVL72 rack
72 NVIDIA Blackwell Ultra B300 GPUs
Warm-water range
36 NVIDIA Grace processors
Supported water types
135 kW design power
HBM4 memory
155 kW peak power
NVLink bandwidth
Exploded view of the NVIDIA GB300 NVL72 rack
72 NVIDIA Blackwell Ultra B300 GPUs
Warm-water range
36 NVIDIA Grace processors
Supported water types
135 kW design power
HBM4 memory
155 kW peak power
NVLink bandwidth

The rack, itemized

SUBSYSTEM

SPECIFICATION / CONFIGURATION DETAILS

Processor

36 × NVIDIA Grace processors per rack

Operating System

Ubuntu 22.04 LTS; RHEL 9.6; OpenShift AI

Chipset

N/A

Accelerators / GPU

72 × NVIDIA Blackwell Ultra B300 GPUs per rack

Memory

960 GB LPDDR5X per tray; 17.28 TB per rack; HBM3e

Primary Storage

Up to 8 × E1.S PCIe 5.0 x4 NVMe SSDs

Secondary Storage

1 × internal 1.92 TB M.2 PCIe 4.0 NVMe SSD per compute tray for OS boot.

Security

TPM 2.0; Secure Boot; firmware resiliency; XClarity security

Chassis

48U rack; 18 compute trays; nine NVLink Switch trays

Power Supply

6 or 8× 33 kW shelves; 135 kW design; 155 kW peak

Ports

Per tray: 4× OSFP; BMC RJ45; USB; Mini DisplayPort

Slots / PCIe Architecture

3× PCIe x16 per compute tray

Bezel

N/A

Form Factor

48U rack-scale system

Network Card

Two ConnectX-8 boards with four OSFP ports; BlueField-3

Warranty

Three-year Premier next-business-day support

NVIDIA rack in a darkened data hall

Infrastructure designed to scale with the rack.

Up to eight 33 kW power shelves

Independent power delivery is sized for sustained rack-scale operation and serviceable redundancy.

Scale-out fabric choice

ConnectX and BlueField networking connect the NVLink domain to high-performance Ethernet or InfiniBand fabrics.

Integrated ConnectX-8 networking

Validated adapters, cabling, and management reduce deployment complexity from rack arrival through operation.

Nuclear-Ready, Plug and Play AI Factories

Bypass grid constraints with self-contained, high-density inference ready infrastructure directly adjacent to clean, base load nuclear power.

Global Deployment Engine

Any Infra, Anywhere in the world.

Global Shipping and Deployment Capability treats advanced infrastructure as mobile, modular freight.

Packaged complex computing nodes, cooling manifolds, and power distribution units into certified, standard form factors.

360° Service Ecosystem

Unifying Legacy with Next-Gen Compute.

Vertically integrated service ecosystem that serves as an operational bridge between legacy environments and next-generation, AI-native infrastructure.

Finance and Leasing

Capital-as-a-Service for High-Density Compute

Convert upfront capital expenditures (CapEx) into predictable operating expenses (OpEx).

Uvation allows enterprises and sovereign entities to deploy multi-megawatt AI factories and advanced GPU clusters without liquidity constraint.