The design of a data center liquid cooling system directly impacts uptime, efficiency, and scalability. With rack densities pushing 200 kilowatts and artificial intelligence (AI) chips drawing over 1000 watts (W) thermal design power (TDP) ratings, cooling begins at the architecture behind coolant distribution units (CDUs).

How CDUs are configured determines how heat is transferred, where it’s rejected, and what systems support it.

CDU configuration choices affect key thermal management factors, including airflow routing, fluid interface compatibility, space utilization, and alignment with site-specific cooling infrastructure. Selecting the right architecture starts with understanding how each one manages heat and what that means for scale, density, and infrastructure planning.

Types of heat rejection methods

Liquid-to-liquid CDUs

Liquid-to-liquid CDUs transfer heat from IT equipment to the facility’s chilled water supply (FWS) through a plate heat exchanger. A dedicated secondary loop or technology cooling system (TCS) circulates coolant through server cold plates, absorbing heat before returning it to the CDU. The two loops stay hydraulically isolated, preventing fluid mixing. Liquid-to-liquid CDUs require chilled water infrastructure to properly remove the heat. Data centers without existing systems may need significant plumbing and pump installations to effectively deploy liquid cooling.

Liquid-to-air CDUs

Liquid-to-air CDUs disperse heat through a heat exchanger coil system and use fans to cool the fluid, allowing operation regardless of access to a chilled water system. This simplifies installation and cuts initial costs by leveraging existing air-cooling infrastructure. However, it is important to note that liquid-to-air and liquid-to-liquid CDUs exhibit distinct thermal performance characteristics. These differences must be assessed against the specific operational requirements of each application for accurate and supportable performance comparisons.

Liquid-to-refrigerant CDUs

Liquid-to-refrigerant CDUs use direct expansion (DX) refrigerant-based technology to remove heat from the data center. These systems are designed to support modular installations, enabling operators to deploy liquid capacity when and where they need it while maximizing efficiency through pumped refrigerant economization.

Specialized components like refrigerant pumps and compressors are needed, along with compliance with environmental regulations for refrigerant leak prevention. The Vertiv CoolPhase CDU was specifically designed to work with the Vertiv Liebert MCV condenser and should be deployed in coordination with system-level design requirements.

Comparing CDU configurations

In-rack CDU

In-rack or rack-mounted CDUs deliver localized cooling by embedding the CDU within the rack. This enables rack-level thermal control, easier deployment in edge sites or retrofits with limited facility liquid infrastructure, and reduces the impact of potential cooling outages to one rack instead of multiple. This type of design reduces system dependencies but adds mechanical complexity inside the rack, potentially affecting cable management, airflow, and serviceability. Limited capacity also makes this setup less suited for larger-scale deployments.

In-row CDU

In-row CDUs centralize cooling between racks, balancing rack-level control with shared infrastructure. They reduce the number of CDUs while enabling scalable liquid distribution. This configuration supports hybrid cooling strategies but requires careful design to avoid thermal imbalance and maintain redundancy. Pipe routing, service access, and failure domain isolation are critical considerations, especially in compact, high-density environments.

Gallery CDU

Gallery CDUs are facility-scale systems positioned at the perimeter or in mechanical corridors. They provide centralized thermal management for large, dense deployments with advanced monitoring and coordinated distribution. These systems support dynamic load balancing and full integration with facility controls. However, they require extensive planning for layout, piping, and control logic – and assume higher operational maturity to manage system-wide cooling at scale.

Wrapping up liquid cooling architectures

Heat rejection strategies vary in complexity, scalability, and energy performance, depending on the design and deployment model. CDU configurations scale differently and suit different floor layouts. Cooling design should align with current site conditions and operational needs, particularly in environments with elevated thermal demands.

Design for density, plan for growth

Find the cooling solution that fits your data center’s needs, with the right design, support, and expertise to match.

Understanding coolant distribution units (CDUs) for liquid cooling.

Contact Vertiv today to discuss your data center cooling requirements.