With AI putting more pressure on data centres than ever, the issue of how these are cooled is also important – and this can create problems for businesses, which resellers need to be aware of.
The growth of AI is revolutionising how businesses work, but it requires a huge amount of computing power to do this. This, in turn, is putting pressure on data centres and means they are generating more heat than ever and how they are cooled needs to be addressed to ensure efficiency and maximum uptime.
Tim Mitchell from Klima-Therm says that AI is fundamentally changing the landscape for data centres. “Unlike traditional workloads, AI processing is incredibly energy-intensive and pushes servers to run at much higher power densities,” he explains. “The International Energy Agency predicts global electricity use from data centres will more than double to around 945 terawatt-hours by 2030, with AI being one of the biggest drivers.
“Right now, AI workloads already account for about 20% of global data centre electricity use, and in some scenarios could reach almost 50% by 2025. That kind of growth puts immense pressure on facilities to maintain uptime while dealing with unprecedented levels of heat output.”
Jeremy Isaacs, sales director – Infrastructure, Cloud & EUC, Ricoh UK, agrees that the growth of large language models and generative AI is driving unprecedented demands on computing power, storage and networks, putting data centres under immense pressure. “This surge in demand has meant that power and heat production have become the limiting factors,” he says.
“At current growth rates, AI technology could drive demand beyond grid capacity within the decade, creating not only operational risks but also fundamental questions around sustainability. The recently signed technology prosperity deal highlights how public and private sectors are beginning to align on this challenge, seeking to balance innovation with resilience, sustainability and equitable access to the digital economy.
“Only through innovation, smarter infrastructure, efficient design and sustainable approaches will organisations be able to unlock the full potential of AI while staying ahead of these pressures.”
VimalRaj Sampathkumar, UK technical head at ManageEngine, agrees that AI is putting data centres under immense strain. “Operators are already reporting unprecedented demand for high-density racks, which significantly raises power draw and the heat that needs to be managed.
“Cooling is now one of the most critical challenges. High-density workloads generate far more heat than traditional ones, and all of it must be managed continuously. Many legacy cooling systems were never designed for this intensity, which leaves operators facing efficiency losses, higher costs, and in some cases real risk of downtime. One third of a data centre’s energy bill can be cooling alone, so the stakes are high.”
Jeremy notes that conventional air-cooling was designed for racks drawing 10–15kW. “In contrast, AI workloads now frequently demand 30–80kW per rack, with some exceeding 100kW,” he adds. “Resulting in overheating, throttled performance and significant efficiency losses across legacy environments.
“Most existing systems are not prepared to handle the sustained thermal demands of AI. To unlock AI at scale, operators must rethink their approach, exploring liquid-assisted and direct liquid cooling as essential strategies to maintain performance, reliability and efficiency.”
Options
There are options for finding the optimal cooling strategy for data centres. “The best option will depend on the scale of the data centre, the age and efficiency of the equipment in use, and the workloads it supports,” notes Markus Rex, general manager at SYNAXON Services.
“Advanced air cooling with hot/cold aisle containment is still widely used but quickly reaches the limits of its capabilities with AI. Liquid cooling is much more efficient at removing high-density heat loads and it’s increasingly being adopted in next-generation facilities.
“Many data centres are taking a hybrid approach, combining air cooling for standard servers with liquid cooling for high-density racks. This may be the best mid-term option in terms of cost efficiency.
“Where sustainability and environmental commitments are a priority, systems can be designed to make use of renewable energy and to re-use heat generated by the data centre.”
Tim says that air cooling alone is not enough in most high-density environments. “The most effective solutions are liquid-based – whether direct-to-chip, cold plate, or full immersion systems. These approaches are far more efficient at capturing and dissipating heat directly at the source,” he says.
“Hybrid systems are also gaining traction, where traditional chilled water systems are supplemented with liquid cooling at hotspots.
“Whatever the cooling mechanism, the by-product is heat, which need not be wasted. This heat can be captured and recycled using district heating systems adjacent to urban developments, including housing developments, manufacturing facilities, hospitals and leisure centres. As well as helping with the decarbonisation effort, this approach also uses less energy than releasing heat into the atmosphere, improving the efficiency of the data centre and lowering bills.”
VimalRaj agrees that liquid cooling is proving to be the most effective route, particularly for AI-driven workloads. “By transferring heat away directly from the chip, it avoids the inefficiencies of simply pumping more cold air through a hall,” he says.
“AI itself also has a role to play, with intelligent algorithms already being used to monitor conditions, predict workload surges, and dynamically adjust cooling in real time. Some providers are pairing these methods with geographic strategies, such as locating facilities in cooler climates to reduce reliance on mechanical cooling.”
Simon Harris, director of critical infrastructure at BCS, notes that while liquid cooling is often positioned as a key solution, its adoption has been relatively slow, due to high initial costs, lack of universal standards and greater familiarity with air-based cooling systems.
“Finally, not all workloads benefit from the high-density cooling capability that liquid systems offer and for lower-density applications, traditional air cooling is often considered sufficient,” he adds.
“But liquid cooling is significantly more efficient at transferring heat, allowing for higher server densities and reduced energy consumption, lower energy bills and operating costs. It is currently the best option for cooling AI and other compute-intensive workloads. It’s also a good option in space-constrained environments, delivering more efficient use of rack space and reducing the overall footprint of the data centre.”
Refurb or rebuild
Which method of cooling will then inform the next steps for what a business does with its data centre(s), which includes a range of options from refurbishment to decommissioning and rebuilding. “Each option has its own challenges, all of which are influenced by cost implications and the sustainability agenda,” says Simon.
“Retaining and upgrading existing stock has several advantages not least that several legacy sites are located on prime land with good power connections that can facilitate extra cooling.
“Most legacy sites can be retrofitted with the specialist equipment needed but it is likely that air cooling will still be part of the solution for the site. Connecting all players in the refresh cycle is key, especially with liquid cooling deployment and its relative lack of maturity in the value chain. Building some partnerships between the client, the technology vendor, the cooling technology specialist and the designers etc will help to create the optimal engineered solution. Objectives must be efficiency, hitting a viable cost point, the programme, the continuity of service for incumbent customers and workloads and future maintenance requirements. This means assembling the right team of client sponsor, professional team, contractor and supply chain.
“Refresh projects are not as straightforward as new builds and require different skill sets. Rather than total rebuilds, many operators are adopting phased upgrades, pairing AI-ready modules or adjacent high-density segments alongside legacy systems maintaining continuity while upgrading capacity. Solutions like liquid cooling retrofits, enhanced containment and reinforced cabinets enable targeted enhancements in existing facilities.”
Sue Preston, vice president, WW Advisory & Professional Services at HPE, adds that for data centre operators to be able to provide infrastructure capacity growth options to meet current demand they require different options. “They can opt to build new data centres able to run modern infrastructure with high density racks and new cooling technologies such as DLC or adaptive cascade cooling (a new patent from HPE) or retrofitting existing data centres to reach the same objective or eventually offloading those new AI use cases to specialised colocation players,” she says.
“One option for new builds are so-called modular data centre solutions, for example, the HPE AI Mod Pod. These offer key advantages as they are prefabricated, pre-engineered components, and in the case of the HPE AI Mod POD feature DLC and a heat-recovery module by design. This not only allows for rapid deployment and shorter construction times but also makes it easier scale data centre infrastructure as needed.
“Adaptive cascade cooling architectures, on the other hand, provide an opportunity for operators that are looking to host new, high-density systems at existing facilities that are not originally fitted for DLC technology. It is particularly effective when retrofitted to existing air-cooled data centres as it can adapt to changing cooling demands and can help reduce energy consumption.”
Future
As AI continues to influence greater aspects of how organisations do business, cooling will remain an important issue. Markus notes that regulatory pressure around energy efficiency and sustainability will add further urgency. “Our advice to resellers is to build their expertise in next-generation cooling technologies as quickly as possible, and look to partner with vendors that offer modular, scalable solutions,” he says.
“Once they have sufficient confidence, they should be raising the subject of cooling within wider discussions on data centre modernisation and positioning themselves as long-term partners who can help customers navigate the delicate balance between performance, cost, and sustainability.”
Tim notes that Goldman Sachs recently forecast that global data centre electricity demand could grow by 160–165% by 2030. “Resellers who want to stay ahead need to understand these dynamics, keep up to date with the latest cooling innovations, and partner with manufacturers and supply partners who are pushing the envelope in terms of energy-efficient technology,” he says.
Sue says that large scale AI and HPC computing deployments, and the inherent question of gaining broader control over costs, operations, resources and cooling will be a key challenge for all data centre operators in the coming years. “To be able to provide not just the optimal solutions but also guide their customers in their long-term strategy, resellers first of all need to familiarise themselves with not just the different cooling technologies, but efficiency levers overall, as the cooling requirements of a data centre are not solely based on the hardware deployed, but also how it is used,” she says.
“By leveraging the tools and insights provided by vendors, working closely with them to understand the IT sustainability and data centre modernisation solutions and services provided, resellers can create their own value-added offerings and establishing themselves as a trusted partner in a field that will see a clear increase in demand.
“These can range from advisory consulting to assess current state and define data centre transformation strategies and roadmaps, to data centre design and build, as well as data centre management services.”
This article first appeared in News in the Channel magazine issue #33.
