Cooling the AI Factory as One System

AI factory performance does not begin and end with the GPU. It depends on how compute, networking, power, cooling, facilities and software work together. As deployments grow denser and larger, the connections between those systems become increasingly important.

That whole-system approach is at the heart of the NVIDIA DSX AI Factory Platform. NVIDIA DSX unifies AI factory design and operations across compute, networking, power, cooling, facilities and software. It helps partners design and operate the factory as one system to maximize AI output reliably within available power, cooling, water and grid constraints.

LiquidStack, a Trane Technologies company, is proud to be at the forefront of that approach. Our GigaModular CDU has qualified as an NVIDIA DSX Ready CDU, meeting the applicable NVIDIA DSX reference design requirements for liquid-to-liquid cooling infrastructure. The qualification helps AI factory builders identify a trusted cooling building block and evaluate its fit within the wider architecture.

Why DSX Ready matters

Choosing the infrastructure for an AI factory involves a series of connected decisions. A CDU cannot be evaluated only by its headline cooling capacity. Its hydraulic performance, thermal behavior, controls, materials and response to changing operating conditions all affect how it will fit within the wider system.

NVIDIA DSX Ready provides a qualification framework for specific partner products and solutions that meet applicable NVIDIA DSX reference design requirements. For builders, this makes the qualified offering and its scope easier to identify. It can help teams evaluate products, address integration requirements earlier and reduce uncertainty as the AI factory moves from design toward deployment.

The qualification covers applicable hydraulic, thermal and control requirements, including constant differential-pressure and constant-flow performance, thermal performance, flow-sensor accuracy, cold-start behavior, pump failover and group control.

Bringing cooling expertise into the framework

LiquidStack’s involvement began before product qualification. We collaborated with NVIDIA during the development of the CDU requirements supporting the initial DSX Ready program, bringing practical experience from designing and deploying direct-to-chip liquid cooling for high-density infrastructure.

That collaboration matters because cooling requirements have to reflect the conditions operators will encounter in real deployments. Flow and pressure have to remain meaningful at system level. Controls must coordinate equipment rather than add complexity. Redundancy and serviceability have to be considered alongside thermal performance. The resulting framework gives the industry a clearer basis for evaluating the role of the CDU within the complete AI factory.

A platform designed to scale

GigaModular is a centralized, scalable CDU platform designed to expand cooling capacity as compute demand grows. Rather than creating a fragmented collection of standalone CDUs, it provides a coordinated approach to cooling across multi-megawatt deployments. 

Its pay-as-you-grow architecture enables operators to align cooling investment with compute deployment. Centralized controls, flexible distribution and configurable redundancy help maintain system-level coordination across racks, rows and data halls. 

This system-level design aligns with the NVIDIA DSX approach, which treats the AI factory as one interconnected system. GigaModular supports that principle through a cooling architecture designed to grow with the wider infrastructure, rather than through a collection of isolated products.

From qualified product to complete delivery

Product qualification is an important starting point. Delivering AI cooling at scale also requires the manufacturing capacity, deployment expertise and service support to support the system throughout its operational life.

As a Trane Technologies company, LiquidStack combines specialist liquid-cooling engineering with broader thermal-management expertise, manufacturing scale and global service capabilities. Together, we support customers as they move from AI factory design and equipment selection into deployment and lifecycle operation.

AI infrastructure is moving toward complete systems designed and operated as one. The qualification of GigaModular as an NVIDIA DSX Ready CDU gives AI factory builders a clearer way to identify liquid-cooling infrastructure that supports that approach.