From Air to Liquid: The Evolution of Data Centre Cooling to Support AI Workloads

Data centres have become the cornerstone of the digital economy, underpinning the rapid evolution of artificial intelligence (AI). As AI workloads intensify, so too does the debate around sustainability and energy efficiency. Today’s largest data centres can consume over 100 MW — enough to power 80,000 homes — making the quest for sustainable operations more urgent than ever.
 

Rethinking cooling for the AI era

Traditional air-cooling methods, long the industry standard, are reaching their limits in meeting today’s high-density, high performance computing environments. The rapid rise of AI and advanced workloads has brought soaring power densities and heat output, requiring a new approach to thermal management. Liquid cooling – once a niche solution – has quickly become essential, offering superior heat transfer capabilities and the potential to dramatically improve energy efficiency.

As AI workloads intensify, liquid cooling offers a path for data centre operators to grow sustainably, with many deploying a blend of air and liquid cooling, optimising for both performance and environmental responsibility.
 

A closer look at liquid cooling solutions

• Direct-to-Chip (D2C) Cooling 

   

  The most prevalent form of liquid cooling circulates a liquid coolant through cold plates attached to processors. This method efficiently draws heat away from the hottest components, reducing the energy required for cooling and enabling higher-density deployments since liquid transfers heat away more efficiently than air. A key component in D2C systems is the Coolant Distribution Unit (CDU), which acts as an isolated loop that ensures the precise delivery, pressure and temperature control of the liquid cooling system.

  
  D2C cooling requires dedicated infrastructure to create this fluid cooling loop, and can be complex to retrofit in legacy facilities if these environments are not designed to support liquid handling and structural support, piping infrastructure and spatial layout to accommodate CDUs, pumps and cold plate loops.  

   
  

  CDUs circulate liquid coolant to cold plates attached to server components.
   

• Rear-Door Heat Exchangers

   

  Rear-door heat exchangers (RDHX) are integrated directly into server racks to intercept and remove heat as it exits the servers. This results in a dual effect: while delivering targeted cooling at the rack level, the chilled air exiting the units can also help reduce ambient temperatures in the data hall.
  
  By bringing the cooling medium closer to the heat source, RDHX offers an effective middle ground between traditional air cooling and full liquid cooling. For operators managing smaller deployments or mixed-density environments, RDHX as part of a hybrid cooling strategy can enhance thermal performance and support higher-rack densities within a constrained footprint, improving the efficiency of existing infrastructure. It also provides the flexibility to scale cooling capacity in line with increasing compute demands. 
   

• Immersion Cooling 

   

  For even greater efficiency, immersion cooling submerges entire server components in a thermally conductive liquid.
  
  The two principal approaches are:

  • Single-phase immersion:  The coolant absorbs heat and is then pumped away to be cooled externally. 

    

    Servers are submerged in a thermally conductive liquid that directly absorbs the heat, and the heated fluid is transferred to a CDU to be cooled.
     

  • Two-phase immersion: By submerging electronics in a dielectric liquid that boils upon contact with heat-generating components, the resulting vapour is condensed and returned to the liquid state in a passive cycle. This efficiently removes heat from high-density AI servers.

   

  Both methods deliver exceptional thermal management, quieter operations and reduced cooling costs compared to air cooling.  However, immersion cooling also represents a transformative shift in data centre design, requiring significant change to server architecture, rack configuration and overall facility layout – often involving large tanks of liquid and customised hardware, and significant upfront investment and considerations to deploy at scale.
   

• Hybrid Cooling Solutions

   

  Hybrid systems combine the strengths of air and liquid cooling, integrating multiple technologies to offer a flexible and scalable approach to managing diverse thermal demands in environments running AI workloads. Liquid cooling technologies effectively target high-density hotspots, while air cooling manages residual heat.

   
  

  Air-based Computer Room Air Conditioning (CRAC) units are combined with liquid-cooled CDUs to deliver adaptable, efficient cooling.
   

This diversified approach allows operators to incrementally enhance their cooling capacity and adapt to varying workload densities, which can enable a smoother transition into supporting high-density AI applications – particularly in currently operational data centres where large-scale infrastructure changes may be challenging. 
 

Unlocking space and performance

Beyond energy and environmental benefits, liquid cooling delivers a practical advantage: space efficiency. By enabling denser computing environments, it can reduce the physical footprint of cooling infrastructure by up to 60%. This allows data centres to pack more compute power into every square metre, supporting business growth without expanding their physical presence.
 

The next frontier in data centre optimisation

What’s clear is that liquid cooling, in one form or the other, will become a mainstay in AI-ready data centres. Escalating power density and heat generation from AI workloads will drive the need for more effective thermal management. The shift towards liquid cooling is not merely a technical upgrade; it is a strategic move towards a more sustainable digital future.

Looking ahead, the relationship between AI and data centres is set to become even more symbiotic. AI is not just driving demand for more powerful infrastructure – it is also poised to transform how data centres are managed, with AI-driven cooling management systems analysing real-time performance data and dynamically adjust cooling strategies to deliver optimal efficiency with minimal environmental impact.

As AI workloads continue to push the boundaries of what is possible, the data centre industry is responding with innovation and resolve. By integrating cutting-edge liquid cooling technologies and harnessing the power of AI for operational optimisation, operators like STT GDC are setting new standards for performance, sustainability, and responsible growth.