Blackwell Generation
NVL72-Class Rack-Scale AI System

GB200-Class Infrastructure Readiness

Caprelion does not sell GB200 systems. Caprelion prepares the complete dedicated physical infrastructure layer for customers deploying GB200-class rack-scale AI servers.

You bring the GB200 NVL72 cluster. Caprelion prepares the dedicated AI POD infrastructure.

AI Rack Power

~130 kW

Planning envelope

GPU Count

72

Blackwell GPUs per NVL72

CPU Count

36

Grace CPUs per NVL72

Cooling

Liquid

Direct-to-chip primary

Infrastructure Specification

GB200 NVL72 — AI POD Infrastructure Planning

System Generation

Blackwell

System Class

GB200 NVL72

Deployment Category

NVL72-class rack-scale AI system

Caprelion Infrastructure Model

Complete SU-class AI POD block

AI Rack Power Envelope

Approx. 130 kW-class (planning assumption)

Primary Cooling

Direct-to-chip liquid cooling

Secondary Cooling

Air cooling support for residual thermal load

Network

High-bandwidth InfiniBand / Ethernet cluster networking

Optical

Optical fiber infrastructure within and between AI POD zones

Power Architecture

Standard AC power distribution with UPS, busway, RPP, PDU

Deployment Model

Complete dedicated AI POD environment — not rack-by-rack colocation

Caprelion Role

Power, liquid cooling, network readiness, site conversion, OEM/EPC coordination, commissioning, operations readiness

Planning envelope only. Subject to site conditions, OEM/EPC validation and final engineering design.

Infrastructure Implications

What GB200 NVL72 Demands

Rack-Scale Form Factor

A single GB200 NVL72 system occupies a full rack-scale form factor combining 36 Grace CPUs and 72 Blackwell GPUs in a single NVLink domain. Infrastructure must be planned at the AI POD / SU level.

~130 kW Per AI Rack

At approximately 130 kW per AI rack, GB200 NVL72 deployments demand robust power distribution. Busway, RPP and PDU infrastructure must be coordinated with the liquid cooling plant from design stage.

Direct-to-Chip Liquid Cooling

GB200 NVL72 requires direct-to-chip liquid cooling as the primary thermal management approach. CDU, XDU, manifold and external heat rejection infrastructure must be planned as part of the AI POD delivery.

High-Bandwidth Cluster Networking

InfiniBand and/or Ethernet cluster networking, plus optical fiber infrastructure, must be planned as part of the AI POD delivery for GB200 NVL72-class clusters.

Commissioning and Operations

AI POD commissioning and handover covers power, cooling, network and monitoring systems. Operations readiness documentation and support are part of the AI POD delivery.

Upgrade Path to GB300

Caprelion AI POD infrastructure for GB200 is planned with upgrade headroom toward GB300 NVL72 (~150 kW). A clear modernization path is part of the design logic.

Complete SU / AI POD Delivery Model

Caprelion does not provide rack-by-rack colocation for GB200 NVL72 deployments. Each GB200 NVL72 cluster is deployed in a complete, dedicated AI POD / SU-class infrastructure block covering power readiness, liquid cooling readiness, network readiness, site conversion, OEM/EPC coordination, commissioning and operations support.

Dedicated AI POD zone — not shared colocation
Power readiness — grid, UPS, busway, PDU
Liquid cooling — CDU, XDU, manifold, heat rejection
Network / optical readiness
Commissioning and handover framework
Compliance-ready delivery
Operations readiness
Upgrade path planning

Infrastructure Evolution

GB200 → GB300 Infrastructure Evolution

GB200 NVL72 — Blackwell

~130 kW

Per AI rack planning envelope. Liquid cooling readiness required. Standard AC power architecture.

Direct-to-chip liquid cooling
Standard AC distribution
CDU / manifold required
8 racks per SU-1.2

GB300 NVL72 — Blackwell Ultra (Current Priority)

~150 kW

Per AI rack planning envelope. Liquid cooling dominant. Higher power uplift vs GB200. Enhanced CDU/XDU.

Liquid cooling dominant
Higher power envelope (+~15%)
Enhanced CDU / XDU
Stronger heat rejection required

Compliance Framework

Compliance-Ready AI POD Delivery

Caprelion AI POD infrastructure is delivered under a compliance-ready framework, designed for alignment with leading data center, security, quality and sustainability standards.

EN 50600

Data Center Infrastructure

Design and operations standard

ISO/IEC 27001

Information Security

Management system readiness

ISO 9001

Quality Management

System readiness

BREEAM

Sustainability Principles

Environmental assessment

CxA

Commissioning Framework

Handover and testing

OPS

Operations Readiness

Ongoing management

Designed for compliance alignment and project-specific certification readiness. Final certification depends on project scope, audit, commissioning, local authority requirements and assessor review. Caprelion does not claim formal certification unless separately confirmed.

Discuss GB200-Class Deployment

Ready to discuss your GB200 NVL72 infrastructure requirements? Contact Caprelion to begin the feasibility review.