CHx2000: The AI CDU

Overview

AI workloads are driving a rapid increase in compute density, pushing cooling infrastructure into new territory. As processors grew hotter and clusters grew larger, traditional CDU designs were no longer sufficient. In response, CoolIT Systems developed the CHx2000—a liquid-to-liquid CDU designed specifically for ultra-dense AI compute clusters.

CHx2000: Maximum Thermal Headroom for Next‑Gen AI

In late 2023, CoolIT Systems recognized that the rapid rise of AI computing would require a new generation of coolant distribution units (CDUs). Semiconductor roadmaps showed GPUs and AI accelerators with ever greater thermal design power (TDP), while hyperscalers shared plans for massive, tightly integrated AI compute clusters. With thousands of CDUs deployed since 2012, CoolIT could see that AI systems would demand cooling at an entirely new scale.

AI workloads, particularly large language models, require massive parallel processing and extremely high data exchange between processors. This drives densely packed designs across processors, servers, racks and entire clusters. Liquid cooling was the clear solution, but the unprecedented density and scale of AI infrastructure required a new generation of DLC systems.

That realization led to the development of the CHx2000. The AI wave would also demand something else—unprecedented manufacturing scale and global deployment support. Backed by its new owners KKR and Mubadala, CoolIT launched an aggressive program to develop a new generation of AI CDUs while rapidly expanding production capacity. Just ten months after concept design began, the first units rolled off the production line at CoolIT’s new Starfield facility in October 2024, built for multi-gigawatt production scale.

The new product line included the AHx240 liquid-to-air CDU for facilities without available water, and the CHx2000 liquid-to-liquid CDU. The CHx2000 quickly emerged as the centerpiece—engineered from the ground up for ultra-dense GPU clusters while providing the scale and global deployment support demanded by AI developers and hyperscale cloud providers.

Engineered for Next‑Gen AI Clusters.

The CHx2000 is defined by its performance. A single CHx2000 can cool up to 12 NVIDIA GB300 NVL72 racks. CoolIT developed CHx2000 with next gen AI in mind, it has the thermal headroom and high flow rates at AI pressure heads to ensure it will future, high-density AI racks.

CHx2000 stands out across three key performance vectors: cooling capacity, flow rate, and pressure head.  These parameters determine whether coolant can circulate effectively through dense server loops and remove heat from thousands of accelerators operating simultaneously.

Group control functionality enables coordinated behavior across up to 20 CDUs, ensuring consistent coolant temperature and flow across large rack groups.

When CoolIT announced the CHx2000 in April 2024 it outperformed alternative CDUs across all metrics. Even in 2026, when some competitors are announcing CDUs with >2MW capacity, the CHx2000 remains the highest density in its class CDU with over 2MW of cooling 750 mm x 1200 mm footprint. While some newer large-sized CDUs are designed for the facility room or grey space, CHx2000 is designed for any space – grey or white.

Simple Integration and Serviceability for Operators

Modern AI clusters demand seamless integration between facility infrastructure, coolant distribution, and rack-level hardware. The CHx2000 is designed to simplify that complexity.

Built on CoolIT’s end-to-end DLC expertise, the system integrates pumps, filters, sensors and controls engineered for tightly coordinated AI cooling environments. Hot-swap components and accessible layouts improve serviceability and uptime, while Redfish support enables integration with existing data-center management tools.

CoolIT engineers work closely with operators to design complete DLC environments—from rack loops to TCS piping—ensuring CHx2000 deployments integrate smoothly into both new and existing data centers.

Systems Engineering and On‑Site Support for Mission‑Critical Deployments

Just as important as the hardware is the support required to deploy and operate large AI compute clusters. These systems demand experienced teams to design, install and maintain complex liquid-cooling infrastructure. CoolIT provides direct liquid cooling (DLC) engineering services along with on-site installation, commissioning and maintenance support in more than 80 countries worldwide.

CoolIT offers support through two complementary capabilities: project engineering and on-site technical services. CoolIT’s professional services team designs DLC systems tailored to each customer’s data center and compute configuration, from single in-rack CDUs to large multi-rack AI clusters. Where required, the team also designs and pre-fabricates technology cooling system (TCS) piping and valve assemblies for rapid on-site installation. Complementing this work, CoolIT’s technical support teams provide installation, commissioning and maintenance services across major markets, helping customers deploy and operate large AI clusters with confidence.

Reliable Supply for Rapid Global Deployment

AI deployment timelines are accelerating. Cooling infrastructure must be available at the pace operators build. CoolIT invested heavily in manufacturing expansion to ensure availability.

The CHx2000 benefits from a production capacity of hundreds of units per month, supported by a global supply chain and quality systems built for multi‑gigawatt scale. In production since October 2024, CHx2000 units have already been installed and are operating at major AI operators’ data centers worldwide.

This supply assurance gives AI operators the confidence to scale clusters rapidly and consistently, knowing that proven cooling systems are ready when needed.

Conclusion

The rise of AI has fundamentally changed the requirements for cooling infrastructure. Compute clusters are denser, processors are hotter, and deployments are scaling globally at extraordinary speed.

The CHx2000 was designed with that reality in mind. It delivers the performance required to cool ultra-dense GPU clusters while being available at the scale and with the global support demanded by hyperscale AI deployments.

For CoolIT, the lesson from 2023 was clear: AI required a new kind of CDU. The CHx2000 was built to be exactly that.