GigaModular CDU

Centralized Liquid Cooling for Hyperscale AI

Scale beyond individual units with a centralized, modular platform – delivering system-level scalability, flexible distribution, and a simplified, pay-as-you-grow approach to capacity and cost.

Download Datasheet
Talk to an expert

Liquid cooling at hyperscale – built to expand without compromise.

GigaModular™ is the original modular CDU platform. Purpose-built for hyperscale AI environments, it is designed to support multi-megawatt deployments and align with emerging AI factory architectures. Unlike traditional approaches that scale by adding discrete units, GigaModular delivers true system-level scalability – enabling you to expand liquid cooling capacity seamlessly, without fragmented infrastructure or loss of efficiency. With a centralized control architecture and configurable layouts, it supports flexible distribution and phased deployment aligned to compute growth – while simplifying CapEx and reducing overall footprint. 

Designed to integrate with facility water systems, building management systems (BMS), and data center infrastructure, GigaModular enables a coordinated, system-level approach to cooling at scale.

Platform Highlights

  • Up to 14 MW scalable cooling capacity per GigaModular (15,000 LPM) 
  • Centralized Control Module with two-way MRP ring network for stable and balanced group control
  • Dual-redundant control, power, and communications for continuous uptime 
  • Flexible deployment architecture supporting N+1 / N+2 redundancy 
  • Supports 45°C water for chillerless cooling 
  • Designed for next-generation GPU platforms, including NVIDIA Vera Rubin
Compare CDU Architectures
Find your CDU

Built for hyperscale environments and aligned with emerging AI factory architectures

Centralized control architecture

Operate and manage your cooling infrastructure through a unified, resilient control system with built-in redundancy.

Flexible distribution

Configure layouts to match your facility, enabling efficient cooling delivery across white space and gray space environments – from racks and rows to entire data halls.

Optimized CapEx and footprint

Deploy capacity as needed with a pay-as-you-grow modelthat reduces upfront investment and minimizes space requirements.

Operational simplicity and resilience

Standardized modules and built-in redundancy (N+1 to N+2) simplify service, improve uptime, and reduce operational complexity.

Hyperscale data centers

AI factories and large-scale training clusters

High-density colocation environments 

Scale With Your Infrastructure

phase 1Initial deployment

Start with a single-module deployment to test performance and integration

phase 2Modular expansion

Scale capacity by adding additional cooling modules as demand increases

phase 3Hyperscale operation

Achieve full-scale deployment with centralized control and optimized efficiency

illustration-deployment phases-defaultillustration-deployment phases-phase-1illustration-deployment phases-phase-2illustration-deployment phases-phase-3

Featured Resources

Frequently Asked Questions

  • Why Direct-to-Chip Cooling for AI

    Direct-to-chip liquid cooling delivers coolant directly to CPUs and GPUs via cold plates, removing heat at the source without submerging hardware. It maintains standard server form factors, supports retrofit and hybrid environments, and offers the most practical path to high-density liquid cooling for most AI deployments. Some residual airflow is still required for non-chip components, making hybrid architectures common in practice.

    Learn More
  • How to Choose a CDU Architecture

    Selecting the right CDU architecture requires evaluating four key factors: rack density, deployment scale, facility constraints, and growth strategy. This guide provides a practical decision framework that maps those inputs to the most appropriate cooling approach — whether in-row, perimeter, or centralized — and explains how LiquidStack’s CDU-1MW and GigaModular platforms address different architectural needs.

    Learn More
  • Liquid Cooling Architecture Types

    Liquid cooling technologies fall into two primary categories — direct-to-chip and immersion — with five distinct architectures across single-phase and two-phase variants. Each offers different trade-offs in thermal performance, system complexity, and deployment model. This article explains how each architecture works, its strengths, and the environments it is best suited for, helping facility-scale buyers evaluate cooling options at infrastructure level.

    Learn More
  • Direct-to-Chip vs Immersion Cooling

    Direct-to-chip and immersion cooling are the two primary liquid cooling approaches for AI infrastructure. Direct-to-chip works with standard server designs and existing data centers, making it the more practical choice for most deployments. Immersion delivers higher overall thermal efficiency but requires purpose-built infrastructure and a fundamentally different operational model. This article explains the trade-offs to help you determine which approach fits your environment.

    Learn More
  • Compare CDU Architectures for AI Cooling

    Choosing the right CDU architecture — in-row, perimeter, or centralized — directly impacts performance, scalability, and long-term cost. This guide compares each approach across key dimensions: efficiency, deployment flexibility, and facility integration. Use it to align your cooling strategy with your density requirements, facility design, and growth plans before you build.

    Learn More
  • Why Liquid Cooling for AI Infrastructure

    Rising chip power densities and AI workload demands are pushing traditional air cooling to its limits. Liquid cooling offers significantly higher heat transfer efficiency, supports rack densities far beyond air-cooled thresholds, and enables the next generation of GPU deployments. For modern AI infrastructure, liquid cooling is rapidly becoming a baseline requirement rather than an optional upgrade.

    Learn More