In Rack CDU Applications in Liquid Cooling Data Centers

As AI Workloads Grow, Cooling Infrastructure Must Evolve

The rapid adoption of AI training, large language models (LLMs), high-performance computing (HPC), and cloud services has fundamentally changed the thermal profile of modern data centers. While traditional enterprise facilities were typically designed for rack densities between 5 and 15 kW, today's AI servers equipped with high-performance GPUs frequently push rack power beyond 30 kW, with many deployments reaching 60 kW or even exceeding 100 kW per rack.

At these power levels, conventional air cooling begins to encounter physical limitations. Increasing airflow, lowering supply air temperatures, or adding more CRAC and CRAH units often results in higher energy consumption without proportionally improving cooling performance. Hot spots, thermal throttling, and uneven airflow become increasingly difficult to manage.

For many existing facilities, however, the primary chilled water infrastructure remains fully capable of supporting higher cooling capacities. The challenge is no longer the building cooling plant—it is how to efficiently transfer cooling capacity from the facility infrastructure to liquid-cooled IT equipment.

This is where the In Rack CDU (Coolant Distribution Unit) plays a critical role.

An In Rack CDU establishes a secondary liquid cooling loop inside or adjacent to the server rack, providing precise temperature control, coolant circulation, pressure management, and continuous monitoring for direct-to-chip liquid cooling systems. Rather than replacing existing facility infrastructure, it enables data centers to integrate liquid cooling while maximizing the value of their current chilled water systems.

This article explores three common deployment scenarios where an In Rack CDU provides practical engineering advantages.

Scenario 1: Retrofitting Existing Air-Cooled Data Centers

Extending the Life of Existing Infrastructure

Many enterprise data centers built over the past decade continue to operate with reliable chilled water plants, cooling towers, and mechanical systems. In most cases, these assets still have sufficient cooling capacity to support future workloads.

The real limitation often lies inside the IT space.

Traditional air cooling relies on moving large volumes of conditioned air through server racks. As rack density increases beyond 20 kW, maintaining acceptable inlet temperatures becomes increasingly difficult. Operators frequently compensate by increasing fan speeds or lowering supply air temperatures, both of which increase energy consumption while providing diminishing returns.

Instead of rebuilding the cooling plant, many organizations are adopting a hybrid cooling strategy.

High-heat components such as CPUs and GPUs are cooled using direct-to-chip liquid cooling, while power supplies, storage devices, networking equipment, and other lower-power components continue to rely on conventional airflow.

This approach minimizes infrastructure disruption while significantly increasing rack power capacity.

How an In Rack CDU Supports Retrofit Projects

An In Rack CDU connects the facility's primary chilled water loop to a dedicated secondary coolant loop serving liquid-cooled servers.

A typical system consists of:

Building chilled water system

Plate heat exchanger

Secondary coolant circuit

In Rack CDU

Rack manifold

Cold plates installed on CPUs and GPUs

The CDU continuously circulates coolant through the secondary loop while isolating server coolant from the building water supply. This separation improves system reliability by protecting sensitive IT equipment from contaminants, corrosion, and water quality variations that may exist in facility piping.

Modern In Rack CDUs also regulate coolant temperature, maintain flow stability, monitor system pressure, and provide leak detection, allowing liquid cooling to operate safely within existing data center environments.

Benefits of Incremental Liquid Cooling Deployment

One of the greatest advantages of an In Rack CDU is that liquid cooling can be introduced gradually.

Rather than upgrading an entire facility at once, operators can deploy liquid-cooled racks only where high-density workloads require additional cooling. Existing air-cooled infrastructure continues supporting conventional servers while AI clusters receive dedicated liquid cooling.

This phased deployment strategy reduces capital investment, minimizes downtime, and allows cooling capacity to scale alongside computing demand.

Scenario 2: AI Data Centers and High-Density GPU Clusters

Cooling Challenges in AI Infrastructure

AI computing places unique thermal demands on modern data centers.

Unlike conventional enterprise servers, AI platforms concentrate enormous heat loads within GPUs, CPUs, HBM memory, and voltage regulation modules. These components generate extremely high heat flux over relatively small surface areas, making efficient heat transfer essential for maintaining performance.

Direct-to-chip liquid cooling addresses this challenge by transferring heat directly from processors into circulating coolant through cold plates.

However, the effectiveness of the entire liquid cooling system depends heavily on the CDU.

More Than a Heat Exchanger

An In Rack CDU performs far more functions than simply exchanging heat.

It serves as the operational control center for the rack-level liquid cooling system by managing:

• Coolant circulation
• Supply temperature regulation
• Return temperature monitoring
• Pressure control
• Flow balancing
• Air removal
• Automatic coolant replenishment
• Water filtration
• Leak detection
• Communication with BMS or DCIM platforms

These functions ensure that every liquid-cooled server receives stable coolant conditions regardless of workload fluctuations.

During AI model training, servers often operate at sustained maximum utilization for days or weeks. Maintaining consistent coolant flow and temperature prevents thermal throttling, allowing GPUs to sustain peak computational performance throughout extended training cycles.

Supporting Future GPU Platforms

As AI hardware continues evolving, power densities will continue increasing.

Rack-level liquid cooling provides a scalable foundation for future GPU generations because cooling capacity can be expanded by adding additional liquid-cooled racks rather than redesigning the entire mechanical infrastructure.

This modular architecture makes In Rack CDUs particularly attractive for AI factories, hyperscale computing environments, and enterprise AI clusters where computing demand continues to grow rapidly.

Scenario 3: Modular Cooling for Small and Medium-Sized Data Centers

Building for Growth Instead of Maximum Capacity

Not every liquid-cooled data center consists of hundreds of racks.

Many organizations deploy smaller AI environments supporting research laboratories, private AI infrastructure, edge computing sites, engineering simulations, or enterprise GPU clusters.

These projects typically prioritize rapid deployment, predictable investment, and future scalability.

Building a centralized liquid cooling system designed for future expansion often increases initial project costs unnecessarily.

Instead, many operators choose a modular deployment strategy based on rack-level CDUs.

Flexible Expansion

An In Rack CDU allows cooling capacity to expand alongside IT infrastructure.

For example, an organization may initially deploy eight liquid-cooled racks supported by a single CDU. As GPU capacity grows, additional CDUs can be installed to support new racks without modifying existing liquid cooling loops.

This modular approach offers several advantages:

Lower initial capital expenditure

Faster installation

Reduced project complexity

Easier maintenance

Simplified future expansion

Because each CDU supports an independent cooling circuit, maintenance activities can often be performed with minimal impact on neighboring racks.

This architecture also provides greater operational flexibility for facilities expecting gradual AI infrastructure growth.

Engineering Considerations When Selecting an In Rack CDU

Selecting an appropriate CDU requires evaluating more than cooling capacity alone.

Several technical factors influence long-term system reliability and efficiency.

Cooling Capacity

The CDU should provide sufficient heat rejection capability for both current workloads and anticipated future expansion.

Pump Performance

Pump head and flow rate must match the hydraulic resistance of the liquid cooling loop while maintaining stable coolant delivery under varying operating conditions.

Heat Exchanger Efficiency

High-performance plate heat exchangers improve thermal transfer while reducing temperature differences between primary and secondary cooling loops.

Redundancy

Critical AI infrastructure often requires redundant pumps, dual power supplies, and fault-tolerant control systems to maximize system availability.

Monitoring and Controls

Comprehensive monitoring of coolant temperature, pressure, flow rate, conductivity, liquid level, and leak detection enables predictive maintenance and faster fault diagnosis.

Integration with standard protocols such as Modbus TCP, BACnet, or SNMP also simplifies data center management.

Maintainability

Features including automatic filling, automatic air purging, replaceable filters, and front-access service panels reduce maintenance time and improve operational efficiency.

Lianli’s In-Rack CDU Solutions:

Lianli offers rack-mount liquid-to-liquid and air-to-liquid CDUs with on-demand customization and project retrofit services. Backed by a 3-year full-unit warranty and remote technical support, these units are designed for liquid cooling in IDC and AI computing facilities—suitable for both new builds and retrofits—and help lower PUE to 1.15. As a direct manufacturer, Lianli has over a decade of deep expertise in the liquid cooling industry.

Conclusion

As rack power densities continue to increase, liquid cooling is becoming an essential component of modern data center infrastructure rather than a specialized technology reserved for extreme computing environments.

An In Rack CDU enables organizations to introduce direct-to-chip liquid cooling without replacing existing chilled water infrastructure, making it an effective solution for both retrofit projects and new AI data center deployments.

Whether upgrading a conventional air-cooled facility, supporting high-density GPU clusters, or building a modular AI computing platform, an In Rack CDU provides the thermal control, operational stability, and scalability required for next-generation computing.

By combining efficient heat exchange, intelligent coolant management, and modular deployment, rack-level CDUs help data centers improve cooling performance while preparing their infrastructure for future advances in AI and high-performance computing.