The rapid development and adoption of AI is putting more pressure on data centres – and how they are cooled – than ever before. Increasingly, hybrid cooling options are being adopted to deal with this.
While the use of artificial intelligence (AI) by businesses has mushroomed in the past two years and provided many benefits to them, it has also put more pressure on data centres than ever before to cope with the increase in demand for computational power. Likewise, it has also put more pressure on the cooling solution used.
As Martin Ryder, channel sales director Northern Europe at Vertiv, notes: “Most data centres in existence today weren’t designed to support the high-density requirements of AI workloads or to effectively remove the higher heat output that they generate. The industry is now facing unprecedented demand for new infrastructure solutions to efficiently power, cool and support this next generation of compute.”
Jens Schumacher, managing partner at STX Group, cites an IDC report that predicted power consumption in data centres will double between 2023 and 2028 to reach 857 terawatt hours. “By 2030, data centres could account for 13% of annual global electricity consumption and 6% of the carbon footprint,” he says.
“AI-driven workloads are also leading to the rise of large hyperscale data centres, which have power demands of 100MW or more. According to the IEA, these facilities consume as much electricity annually as 350,000 to 400,000 electric cars.”
Rupert Bedell, managing director at Fasthosts, notes that things like deep learning and machine learning will require significant processing power in order to function properly. “This has led to the integration of GPUs and high-performance computing (HPC) systems within data centres,” he adds. “This shift is pushing the design of more powerful and efficient data system infrastructure.”
Indeed, as Alex Sharp, chief operating officer at Nscale, adds: “Data centres traditionally relied on CPUs for general purpose tasks, but these aren’t suitable for supporting AI workloads,” he says. “Over the past few years, with more and more companies adopting AI, the balance of hardware in data centres has shifted towards GPUs, which are best suited to handle AI workloads. This also means that AI data centres have significantly higher power and cooling requirements compared with infrastructure from just a few years ago.
“Alongside much-increased power requirements, GPUs for AI workloads run hotter than other processors. With 132KW rack power requirements, it is no longer possible to rely solely on air cooling.”
Advantages of hybrid cooling
As Alex notes, with air cooling alone not up to the task, more businesses are looking to hybrid cooling solutions – which use a combination of air and water cooling – to cope.
Martin says that hybrid cooling is a key strategy for infrastructure designed for AI workloads. “Liquid-cooled solutions efficiently remove heat from the GPU servers while air-cooled systems handle the heat from traditional CPU hardware, as well as allow other networking equipment and in rack/wider room considerations to be catered for,” he says. “The diversification of cooling methods in a hybrid approach enhances redundancy and reliability and diminishes the risk of system failures.”
Martin adds that the flexibility and scalability of hybrid cooling allows for tailored solutions. “These can adapt to changing workload demands, facilitating incremental upgrades and expansions as necessary, without a complete overhaul of existing air-cooling systems,” he says.
Jens agrees that adaptability is a key advantage of hybrid cooling. “It enables data centres to optimise cooling strategies in real time, switching between liquid and air cooling based on workload demands,” he says. “This enhances efficiency [and] reduces energy consumption and operational costs.”
Nick Ewing, managing director of EfficiencyIT, says that hybrid cooling solutions offer a balanced approach. “They leverage the cost-effectiveness and simplicity of air cooling while integrating the superior heat removal capabilities of liquid cooling,” he says.
“Hybrid solutions optimise performance and resiliency and ensure consistent cooling for GPU-accelerated computing systems, even under variable loads. They also help maintain operational reliability by using methods such as direct-to-chip liquid cooling alongside conventional systems, allowing end-users to created mixed technology environments.”
Investment
Hybrid solutions also have the advantage of being cheaper than transitioning to a complete liquid cooling solution, which, as Alex notes is an expensive and time-consuming process if you are looking to upgrade and re-tool a legacy data centre. “But hybrid cooling systems allow data centre operators to target the areas that need it most first – the locations with the highest-density compute in a data centre,” he says.
Jens agrees that transitioning to liquid cooling requires substantial investment. “Hybrid cooling solutions present a cost-effective step and strategic immediate step,” he says. “By integrating liquid cooling for high-density racks while retaining air-cooling for lower heat intensive systems, data centres can optimise budgets and achieve improvements in cooling efficiency. This approach allows for gradual adaptation to evolving thermal management needs without the immediate financial burden of a full-scale liquid cooling transition.”
Environmental impact
Another advantage of hybrid cooling solutions is they have less of an environmental impact than traditional methods. Martin says that the evolution of GPU’s mean that greater heat is generated from the servers. “This in turn makes it more flexible to re-use in a wide range of use cases,” he says.
“High-temperature fluid coolers operate efficiently at return and supply water temperatures of up to 50/40°C respectively, enabling data centres to reduce their cooling loads without compromising reliability. This opens the door to circular economy practices like heat recovery.
“By capturing waste heat and reusing it for applications such as district heating or thermally powering nearby industrial processes, operators can reduce their carbon footprint.”
Nick Remington, technical director at Black & White Engineering, says that the specific heat capacity of water compared to air is about four times more effective, enabling hybrid cooling solutions to offer much more efficient solutions, reflected through improved facility Power Utilisation Effectiveness (PUE).
He adds that a hybrid cooling solution compared to traditional methods reduce operational emissions (Scope 2) making less impact to the environment. “Other environment benefits include the opportunity to elevate the Facility Water System (FWS) serving the liquid cooling infrastructure,” he says.
“The higher the operating temperature of the FWS for rejecting heat, means a greater opportunity to reduced chiller compressor operation, further reducing power consumption and lowering PUE further. Liquid cooling infrastructure uses cooling distribution units (CDU) to deliver the cooling to the rack. For the same amount of cooling delivered, the electrical demand for CDU pumps compared to air cooling fan power consumed is significantly lower again assisting with lowering PUE.
“Higher temperature FWS means a greater opportunity for useful heat export to the local surrounding area and potential district heat networks. Utilising heat export offers minor data centre improved efficiencies but has a greater impact on reducing energy demand of buildings served by the district heat network.”
Rupert adds that hybrid solutions can leverage free-air cooling and evaporative cooling, when environmental conditions are favourable. “In traditional cooling methods – such as refrigeration or air conditioning – energy consumption stays at a consistently high level,” he says. “This increases the operational costs, and contributes to the data centre’s carbon footprint, especially as the high electricity usage often comes from non-renewable sources.
“But hybrid solutions incorporate more sustainable techniques to optimise energy usage, lower the need for artificial cooling, and cut down on greenhouse gas emissions. These systems are also often designed with precision cooling that targets the most heat-sensitive areas of a data centre, minimising wasted energy and improving efficiency.”
Nick agrees that the balanced approach of hybrid cooling means less energy is wasted. “The systems can also be fine-tuned to match real-time demand,” he says. “This efficiency is particularly important as the data centre industry seeks to align with more sustainable practices in response to growing environmental concerns around AI and GPU-powered compute.”
Paul Beech, commercial director of Armatherm, adds that by using natural cooling methods when possible, they reduce the need for energy-intensive mechanical cooling, lowering energy consumption and carbon emissions. “Maintaining effective thermal isolation helps cooling systems operate efficiently, minimising the risk of wasted energy and ensuring that cooling efforts are focused where they’re needed most,” he says.
Reseller conversations
When talking to customers about hybrid cooling solutions for data centres, there are various aspects that resellers should highlight. “By combining free-air, liquid and traditional methods, hybrid solutions result in significantly reduced energy consumption and lower operating costs,” says Rupert. “They are also flexible and scalable, making them ideal to handle the growing demands of AI and big data workloads.
“As data centres strive to become more sustainable, hybrid cooling solutions are incredibly attractive as they use fewer resources and produce fewer emissions. Hybrid cooling also offers reliability and redundancy, ensuring continuous cooling even if one method – such as free-air – becomes less effective for any particular reason.
“The future-proof benefits of hybrid cooling should be highlighted, as these systems are easily adaptable to meet increasing data processing demand, making them a smart long-term investment for data centres.”
Jens adds that resellers should emphasise how hybrid systems enhance reliability by mitigating risks. “If one of the cooling methods experiences an issue, the other can help maintain the optimal temperatures, reducing the risk of downtime,” he says. “While liquid cooling is well suited for high-intensity workloads, there are scenarios where air cooling remains more effective, underscoring the adaptability of hybrid solutions to diverse data centre needs.”
Martin adds that resellers should tell customers that they not only need the right technology, but also a partner with a broad understanding of the end-to-end power and thermal chain. “They should be able to offer a broad portfolio of innovative liquid and air-cooling technologies, and a full range of power solutions and services,” he says. “This means working with a vendor that has a large service network and global footprint with the capacity for speed and scale.”
Nick adds that it’s important to stress the advantages of using a unified, turnkey solution over assembling components from multiple vendors. “This is a common pitfall that can lead to inefficiencies and higher maintenance costs,” he says. “A well-integrated system ensures greater resilience, easier maintenance, and more predictable performance, which is critical for organisations relying on data centres to support mission-critical applications.”
Growing demand
Demand for hybrid cooling solutions are set to continue to grow as the use of AI is expected to expand in the coming years. “The ongoing surge in AI-driven applications is set to push data centre capacities to their limits in terms of computational power and thermal output,” says Nick.
“Hybrid cooling solutions are ideally positioned to meet this challenge because they can be tailored to address the unique requirements of high-performance computing systems, while enabling users to deploy traditional enterprise compute within the same environment. As businesses increasingly invest in AI, the need for scalable, efficient and robust cooling solutions will naturally drive demand for hybrid systems that are capable of future-proofing data centre operations.”
Paul agrees that as AI continues to advance, the demand for high-performance, energy-efficient cooling solutions will rise. “Hybrid systems are likely to become essential in managing the increased heat loads from AI servers,” he says. “Data centres that prioritise thermal efficiency and continuous insulation will be better equipped to meet these demands while staying in control of energy costs.”
Alex notes that data centres will only run hotter as more compute capacity is crammed into a single server rack and eventually air cooling alone will no longer be suitable for keeping chips cool. “Hybrid cooling systems will act as a stopgap for data centre operators looking to improve energy efficiency without needing to retrofit an entire facility,” he says. “However, it’s important to recognise that this is just a temporary solution and that liquid cooling will eventually become the standard.
“We’re already seeing more and more greenfield data centre builds opting for completely liquid-cooled solutions as the initial investment needed can be less than the cost of retrofitting later. This will only continue – hybrid cooling is just a cheaper alternative to retrofitting an entire data centre at once.”
