Data Centers and AI: Liquid Cooling Accelerates in Europe

The rise of artificial intelligence is rapidly changing the way data centers are designed and cooled . Increasing computing power is driving ever-higher thermal densities in racks, making it more difficult to dissipate heat through traditional air systems.

In European data centers, this transformation is accelerating the use of liquid cooling and hybrid architectures , which combine liquid cooling with air systems. The shift isn't limited to the technology installed near the servers; it also involves hydraulic distribution, exchangers, CDUs, chillers, dry coolers, and the overall control and management of the refrigeration infrastructure.

AI racks increase power density

Traditional cloud and enterprise workloads typically operated at moderate densities, in the 8-12 kW range per rack. With AI , however, GPU- and accelerator-based clusters can frequently reach densities of 30-80 kW , with deployments even exceeding 100 kW per rack.

Data collected in the European sector already shows this evolution: a significant portion of operators declare AI racks between 50 and 99 kW, while approximately 15% report powers equal to or greater than 100 kW.

At these levels, simply increasing the air flow becomes less and less effective. In fact, the following increase :

• the amount of air to be moved;
• the power required by the fans;
• the difficulty in maintaining uniform temperatures;
• the risk of hot spots;
• the complexity of air distribution in data rooms.

Liquid cooling, on the other hand, allows heat to be transferred directly from the CPU and GPU to a heat transfer fluid, reducing the thermal path and increasing the amount of heat that can be removed from relatively compact surfaces.

Direct liquid cooling increasingly present

Among the technologies that are growing the most is Direct Liquid Cooling (DLC) , a category that also includes direct-to-chip configurations.

In direct-to-chip, cold plates installed directly on the hottest electronic components transfer thermal energy to a liquid circuit. The heat then passes through a CDU (Cooling Distribution Unit), which separates the IT circuit from the building circuit and controls its flow rates, temperatures, and pressures.

According to data collected from European colocation and hyperscale data center operators, air cooling remains widespread, but approximately 47% of respondents already use DLC solutions for high-power racks. Immersion cooling, on the other hand, remains much less common.

This suggests that the transition won't happen through an immediate replacement of air-based systems. Many data centers are instead adopting hybrid architectures , using liquid for high-density CPUs and GPUs while retaining air for less critical components or for general room cooling.

Liquid cooling also changes the refrigeration system

Introducing liquid near servers shifts some of the complexity from the data room to the building's thermal infrastructure.

A system may include :

• cold plate or immersion modules;
• CDU;
• heat exchangers;
• water or water-glycol pumps and circuits;
• dry cooler or chiller;
• control and supervision systems;
• redundancies necessary to ensure operational continuity.

For the refrigeration sector, this opens up an increasingly important field of application. The project not only concerns the production of cold, but also the management of the entire process by which heat is transferred from the chip to the external environment.

Another potential advantage is the ability to use higher water temperatures than traditional chilled water systems . This can increase the hours during which heat is directly removed via dry coolers or free cooling, reducing the need for mechanical refrigeration.

At the same time, higher return temperatures also increase the potential for heat recovery to buildings, district heating networks, or other users . The European Union is paying increasing attention to the possibility of integrating data centers into local energy systems through waste heat recovery.

Efficiency and reliability become inseparable requirements

However, data center cooling has a peculiarity: efficiency cannot be pursued at the expense of reliability .

IT infrastructures must operate continuously, and any cooling system issue can put extremely expensive equipment at risk. Therefore, the transition to liquid cooling requires continuous monitoring, redundancy, and the ability to respond quickly in the event of anomalies.

Among the parameters to be monitored, the following become central:

• flow and return temperature;
• liquid flow rate;
• circuit pressure;
• quality and conductivity of the fluid;
• any losses;
• CDU performance;
• energy consumption of dissipation systems.

The growth of AI is making cooling an increasingly strategic component of digital infrastructure. According to international estimates, data center electricity consumption could more than double by 2030, primarily due to accelerated computing used in AI applications.

For refrigeration, this means a rapidly evolving market in which traditional skills in heat exchange, hydronic circuits, chillers, and control are increasingly integrated with the specific needs of the IT world.