LiquidStack launches modular CDU for AI data centres
Fri, 22nd May 2026 (Today)
LiquidStack has launched its GigaModular coolant distribution unit platform, which supports deployments of up to 14MW. The system is now commercially available and has already attracted early customer orders.
The launch targets a growing part of the data centre market as operators look for ways to cool denser AI and high-performance computing systems without overhauling site infrastructure in a single step.
Liquid cooling has risen up the agenda as newer AI servers generate more heat than conventional air-cooled systems can efficiently handle at high rack densities. In response, suppliers and operators are focusing on equipment that can be deployed in stages as computing demand grows.
GigaModular uses a modular design that allows operators to add cooling capacity in multi-megawatt blocks rather than installing it all at the outset. The product is designed to meet NVIDIA Vera Rubin specifications.
The platform is aimed at AI, HPC and hyperscale data centres. It is intended to support a range of chip designs and heat profiles across different operating temperatures, including both merchant silicon and custom silicon used by large cloud groups.
Modular design
The system centres on a coolant distribution unit, or CDU, which manages the flow of liquid used to remove heat from servers and related equipment. With a larger CDU platform, LiquidStack is seeking to address one of the practical constraints facing operators as AI deployments move to higher densities and larger clusters.
The expanded platform allows cooling infrastructure to be added in phases. It also includes centralised controls designed to reduce duplication across installations and make operations easier to manage at scale.
LiquidStack said the product has completed multi-module system integration and full-load testing for deployments up to 14MW. The platform has also received ETL certification.
LiquidStack is part of Trane Technologies, which has an established presence in thermal management and building systems. The CDU platform sits within Trane Technologies' wider thermal management offering and can be supported through its broader service network across deployment and equipment life cycles.
Market demand
The commercial release comes as data centre operators and suppliers race to adapt sites for AI workloads, which often require more electrical power and more intensive cooling than traditional enterprise applications. A phased approach to liquid cooling infrastructure can help operators match capital spending to customer demand and reduce the risk of overbuilding capacity before it is needed.
Joe Capes, Vice President at Trane Technologies and General Manager of LiquidStack, outlined the company's position on the launch.
"Purpose-built for high-density AI environments, GigaModular helps operators expand cooling capacity without requiring significant infrastructure redesigns or over-provisioning upfront," said Joe Capes, Vice President at Trane Technologies and General Manager of LiquidStack.
"Its flexible deployment and distribution architecture enables phased buildouts aligned to AI compute demand, while centralized system-level controls help simplify operations, improve efficiency and optimize capital planning," he said.
LiquidStack first announced the GigaModular CDU platform in 2025. The latest announcement marks its move into commercial availability following testing and certification, alongside what it described as initial customer demand.
The platform is intended to scale beyond a collection of separate CDUs by supporting coordinated, system-level expansion. That distinction matters for operators managing larger AI estates, where cooling, electrical design and controls increasingly need to work as an integrated system rather than as isolated components.
Rising demand for AI data centre capacity has increased scrutiny across every layer of site design, from grid connections and backup power to chillers, liquid loops and rack-level thermal management. Suppliers that can offer equipment in standardised blocks may find a ready market among operators trying to bring new capacity online quickly while preserving flexibility to adjust layouts and densities later.
LiquidStack said its design is intended to accommodate changing workloads, rack densities and facility layouts as AI systems evolve. Early customer orders suggest some operators are prepared to back that model as they weigh how to build cooling systems for the next generation of GPU-based infrastructure.