Liquid gold?

As the demand for data processing power increases – most notably with the rise of artificial intelligence, but also thanks to the increasing demands of high-frequency trading, cloud computing, and edge computing - so does the need for efficient thermal management in data centers. Traditional air cooling systems, while still prevalent, are increasingly challenged by the density and heat output of modern computing hardware. As a result, data centers are increasingly turning to liquid cooling technologies for superior performance, energy efficiency, and environmental sustainability.

Why move beyond air cooling?

Before examining specific liquid cooling technologies, it's worth taking a moment to understand the limitations of air cooling, while acknowledging that this longstanding, ‘traditional’ approach will continue to be suitable for many data center workloads.

There are 4 main areas where air cooling’s limitations in coping with modern, high density, AI and HPC applications can be exposed: 

1. air cooling struggles with heat dissipation in high-density server environments (10+ kW per rack);
2. uneven airflow can create localized “hot spots,” increasing the risk of equipment failure; 
3. air cooling systems often require significant energy to power fans, chillers, and CRAC (Computer Room Air Conditioning) units; 
4. and air cooling demands raised floors, ducting, and larger physical layouts to ensure sufficient airflow.

In contrast, liquid cooling leverages the superior thermal conductivity and heat capacity of liquids, offering more targeted and energy-efficient heat removal solutions.

Liquid cooling options

There are 5 main technology solutions which come under the heading of liquid cooling. As of now, there is no clear-cut preferred choice for data center use, although it is likely that, over time, one or two clear winners might emerge, allowing for different liquid cooling approaches best suited to different high density workloads.

1. Direct-to-Chip (D2C) Liquid Cooling

In direct-to-chip cooling, cold liquid (usually water or a dielectric coolant) is circulated through cold plates that are attached directly to key heat-generating components like CPUs, GPUs, and memory modules. The liquid absorbs heat and is then carried away to be cooled and recirculated.

There are several advantages to this approach. Liquid directly absorbs heat from the hottest parts of the server, significantly reducing reliance on ambient air. D2C systems can often be integrated into existing server hardware with minor modifications. Higher rack densities are supported (20–50 kW/rack or more). Fans can be smaller or eliminated, reducing noise and mechanical failure points.

2. Immersion Cooling

Servers or components are submerged in a dielectric liquid that does not conduct electricity. There are two main types:

•  Single-phase immersion: The fluid absorbs heat and is pumped to a heat exchanger to be cooled.
•  Two-phase immersion: The fluid boils when it absorbs heat, changing from liquid to vapor. The vapor is condensed back into liquid and recirculated.

Immersion cooling offers a range of benefits, including exceptional heat transfer (direct contact with components ensures efficient heat absorption), no airflow needed (removes the need for fans and CRAC units), reduced equipment wear (no mechanical stress from airflow or dust accumulation), smaller footprint (high thermal efficiency allows for denser configurations in smaller spaces), and sustainability (can reduce PUE to close to 1.0.)

Set against these are the potentially high upfront costs of new tanks and associated equipment, plus likely hardware adaptation. 

3. Rear-Door Heat Exchangers (RDHx)

Mounted on the back of a server rack, RDHx systems use a liquid-cooled coil to absorb heat from air as it exits the servers. The warm liquid is pumped away and cooled externally.

The advantages of this technology are the fact that it is easy to retrofit, RDHx can be added to existing racks without major design overhauls; there is no direct contact, liquid does not touch server components, reducing perceived risk; and improved energy efficiency, reducing reliance on traditional room-scale cooling.

4. Cold Plate Cooling

Cold plate cooling is a subset of direct-to-chip cooling where metallic plates with embedded channels carry liquid coolant across the surfaces of key components. Unlike immersion, cold plate systems are semi-contained, avoiding liquid spillage risks.

Challenges with this technology include complex plumbing, careful engineering is required to manage tubing, valves and flow paths, along with possible maintenance concerns as there are more moving parts and potential leak points compared to passive cooling.

On the plus side, cold plate cooling is high precision, targeting only the components that require direct cooling, requires lower coolant volume, using less liquid than immersion systems, and offers flexibility, as it is compatible with industry standard rack formats.

5. Liquid-to-Air and Liquid-to-Liquid Heat Exchangers

These systems act as intermediaries between the data center's internal liquid cooling loop and external cooling infrastructure.

•  Liquid-to-Air: Cools the heated liquid using external air (e.g., via dry coolers or cooling towers).
•  Liquid-to-Liquid: Transfers heat from the server loop to a building-wide chilled water system.

The main advantages of these systems is the fact that they can be integrated into existing HVAC infrastructure and they deliver significant energy efficiency, as they optimise the entire heat rejection path.

Liquid Cooling versus Air Cooling

A comparison between air cooling and liquid cooling reveals several comparative advantages for the liquid option. In terms of thermal efficiency, air cooling is limited, especially in dense configurations, where liquid cooling provides high thermal efficiency, especially with immersion or D2C. Air cooling is effective for rack power densities typically between 5-10kW/rack, while liquid cooling can cope with up to 100+ kW/rack (and there is talk of the need to cool 1mW/rack at some stage in the future!). Energy consumption-wise, air cooling is relatively high due to its fans, chillers, CRAC units etc. Liquid cooling has a lower overall energy consumption, especially in closed-loop systems. Liquid cooling is much quieter than air cooling, requires a smaller footprint, offers reduced wear and tear and also a lower PUE.

While the benefits of liquid cooling when compared to air cooling are considerable when it comes to the current focus on AI and other high performance compute workloads, it is worth noting some potential issues which need to be considered as part of any liquid cooling implementation. The initial capital expenditures for liquid systems can be higher than for air cooling solutions, though operating costs are often lower. Additionally, maintenance teams may need retraining to handle liquid-based systems. Not all server hardware is designed for liquid integration - especially immersion-based solutions. Leak management is an important consideration - redundancy and monitoring systems are essential for preventing and detecting leaks.

Conclusion

Liquid cooling represents a paradigm shift in data center thermal management. From immersion cooling to rear-door heat exchangers, the technologies are diverse but united by their ability to outperform air-based systems in high-density, high-performance environments. Though implementation can be complex and capital-intensive, the long-term benefits in energy efficiency, sustainability, and thermal control make liquid cooling a compelling solution for the future of digital infrastructure.