The Rise of Immersion Liquid Cooling: Powering the Next Wave of AI Infrastructure

Artificial intelligence (AI) has moved rapidly from a conceptual technology to a transformative force of industries, reshaping business models and digital infrastructure.  As AI models become more sophisticated, the demand for computational power continues to surge, driving hyperscalers and enterprises alike to seek new solutions for data centre design and operations.

Today, rack power densities in data centres have soared from the traditional 10kW per rack to well over 100kW, stretching the limits of conventional air-cooling systems. In response, beyond other liquid cooling methods such as direct-to-chip, immersion cooling – a process in which servers are submerged in thermally conductive dielectric liquid – has emerged as a compelling solution for next-generation AI workloads. This technology offers superior thermal management, enhanced energy efficiency and reduced maintenance requirements compared to traditional cooling systems.

Why immersion cooling matters for AI
In AI environments where GPU clusters are densely packed and continually running at full throttle, effective heat dissipation becomes critical. Liquid immersion cooling addresses this challenge head-on by enabling direct and efficient heat removal.

1. Meeting escalating thermal demands
   Liquid immersion cooling eliminates the hotspots often found in densely packed graphics-processing-unit (GPU) clusters, as it removes heat from the submerged servers. Modern immersion systems can support cooling capacities exceeding 200kW per rack, allowing operators to scale infrastructure in line with the growing intensity of AI applications.
    
2. Delivering consistent, predictable performance
   By eliminating hotspots and preventing thermal throttling, immersion cooling ensures that processors and accelerators consistently operate at peak performance. This stability is critical for businesses seeking to accelerate product development cycles and maximise model throughput, especially as the complexity and scale of AI models continue to expand.
   In addition, it provides more constant server temperatures than traditional cooling methods, allowing businesses to obtain more predictable performance. This in turn accelerates technology-related product development cycles and boosts model throughput.
    
3. Driving energy efficiency and sustainability 
   Immersion systems can achieve Power Usage Effectiveness (PUE) ratios as low as 1.1 (an ideal PUE is 1.0). These gains translate directly to operational cost savings and reduced environmental impact.
   Notably, immersion cooling reduces the need for water in the cooling process and leverages the superior heat transfer properties of dielectric liquids, enabling efficient operation at higher temperatures. 
    
4. Maximising space management 
   By significantly increasing the power density that each rack or tank can support, immersion cooling helps optimise space management by reducing the overall footprint needed for high-performance workloads. This reduces the need for bulky traditional cooling infrastructure, allowing for compact, modular immersion cooling tank designs that enable operators to maximise floor space while efficiently managing the thermal demands of AI and other compute-intensive applications. 
    
5. Weighing maintenance considerations 
   Immersion cooling offers certain protective benefits – by submerging servers in dielectric fluids, systems are shielded from dust and corrosion caused by humidity. While this can reduce certain routine maintenance tasks, servicing equipment in a liquid environment introduces its own set of operational complexities. This includes having robust leak detection systems in place, regularly monitoring fluid quality to ensure dielectric strength and contamination control, and carefully handling and reintegrating components following maintenance or replacement.  
    

Navigating challenges and considerations  

As immersion cooling gains traction for supporting high-density AI workloads, data centre users—whether enterprises or cloud service providers—should be aware of several important considerations. Understanding these factors will help the data centre user make informed decisions.

1. Hardware compatibility and support
   To get the most out of immersion cooling, it’s important to select IT equipment that’s designed or certified to perform optimally in an immersion-cooled environment. Many hardware vendors offer equipment built with this in mind, enabling better thermal performance and longer component lifespan. Engaging early with your hardware provider helps ensure alignment on specifications and ongoing support, so you can confidently deploy your AI workloads with peace of mind. 
    
2. Service continuity and maintenance
   Immersion cooling can reduce maintenance needs and extend hardware lifespan, but it also changes how equipment is serviced. Data centre maintenance protocols, response times, and hardware swaps or upgrades are managed in an immersion environment are all important.
    
3. Skills and support ecosystem
   Operating in an immersion-cooled environment may require new skill sets and operational processes that data centre providers has trained staff and robust vendor partnerships to support your infrastructure.
    
4. Sustainability and efficiency gains
   Immersion cooling offers significant energy and water savings. Data centre providers can advise on Power Usage Effectiveness (PUE), water usage, and any sustainability certifications, so users can accurately assess the environmental benefits for their organisation.
    

Ready for AI. Ready for what’s next.

More than a response to rising AI demands, immersion cooling represents a step towards reimagining how data centres are designed, built and operated for the future. 

As AI applications continue to evolve, data centre operators must remain agile, designing facilities with adaptability at their core and taking a proactive approach to meet customer requirements. Immersion cooling is more than a technical upgrade; it is a strategic capability that enables operators to meet the demands of today’s AI workloads while remaining flexible for future innovations.

With data centres that are purpose-built for AI workloads and equipped with the latest advanced cooling solutions, users can be confident their digital infrastructure is ready for the future. STT GDC’s AI-ready facilities are designed to support the next generation of high-density, compute-intensive accelerated computing workloads, empowering your organisation to innovate at scale and stay ahead in an increasingly digital world.