Artificial intelligence (AI) enables machines to learn from data, recognise patterns and generate content. As its use expands across business and everyday life, the infrastructure required to train and run AI models is growing rapidly. According to the International Energy Agency (IEA), electricity consumption in AI-focused data centers increased by 50% in 2025.
For colocators, owners and investors, this creates an opportunity to develop new computing capacity, alongside a challenge: making that capacity available efficiently and reliably. Higher rack densities and changing workload profiles place new demands on cooling systems. The architecture selected today will influence operating costs, the use of available electrical power and the facility’s ability to accommodate future IT deployments.
AI workloads rely on powerful accelerators that concentrate substantial computing capacity within individual racks. The IEA reports that AI server power density increased elevenfold between 2020 and 2025. This illustrates the pace at which infrastructure requirements are changing, with implications for both new developments and facilities adapting to AI customers.
Meeting this demand involves more than adding cooling capacity.Developers must coordinate equipment procurement and construction with the availability of electrical power while accommodating IT specifications that may evolve during the project. For a colocation operator, delays can affect when contracted capacity becomes available; for an investor, they can postpone the point at which the facility starts generating returns. Cooling decisions made early in this process can help avoid later changes to layouts, equipment interfaces and installation requirements.
Once operational, AI infrastructure also presents a different load profile. NVIDIA explains that large training clusters can synchronise thousands of GPUs, causing abrupt changes between higher and lower electrical demand. Traditional enterprise workloads are often more diverse and asynchronous, so individual fluctuations are more likely to average out across the facility.
These coordinated variations affect heat generation, although their impact on coolant temperatures depends on the thermal characteristics of servers and the wider cooling circuit. Designers must therefore consider how the system responds over time, alongside the maximum load it can support.
The challenge is to connect these operational requirements with a deliverable project: an architecture that can accommodate high-density IT, manage variable loads and be installed within the required schedule.
Understanding this operating profile changes how cooling performance should be assessed. A system selected only for its peak design condition may perform differently at partial load or during changes in IT activity. The assessment must consider the conditions the facility will experience throughout the year.
Power Usage Effectiveness (PUE) provides a way to evaluate the energy overhead associated with supporting IT operations. It is calculated by dividing total facility energy consumption by IT equipment energy consumption over the same period. A PUE of 1.20 means that every unit of energy consumed by IT requires another 0.20 units for infrastructure such as cooling and electrical distribution.
For a given IT energy requirement, a lower PUE means lower facility electricity consumption. For owners and investors, this makes cooling efficiency relevant to long-term operating expenditure. It can also help make better use of a constrained electrical connection, potentially leaving more power available for computing.
Dynamic performance deserves attention within this assessment. Thermal buffering and coordinated controls can help stabilise water temperatures as IT loads change. Where this avoids inefficient plant operation, it may support better annual energy performance. The effect depends on the complete system and its control strategy.
Temperature stability also supports operational continuity, a benefit that PUE alone cannot describe. Project teams should therefore assess energy performance and the ability to maintain suitable server cooling conditions together, ensuring that efficiency targets are compatible with the availability requirements of the intended AI workloads. This is particularly relevant when commissioning a facility in phases, before its eventual design of IT load has been reached.
These requirements make the choice between air and liquid cooling an important part of the design. Air cooling transfers heat from servers into airflow, which is then handled by equipment such as computer room air handlers or fan wall units. It remains suitable for many installations, but increasing rack densities make it harder to remove heat efficiently using air alone.
Direct-to-chip liquid cooling captures heat at processors and accelerators through cold plates. The heated coolant carries this energy to a Coolant Distribution Unit (CDU), which typically transfers it through a heat exchanger to the facility water circuit. Keeping these circuits separate allows the coolant serving sensitive IT equipment to be managed independently of the facility water.
Discover how the Daikin CDU efficiently manages heat from AI processors
Capturing heat close to its source reduces the airflow required for cooling. Where IT specifications permit warmer coolant temperatures, the facility can also use outdoor conditions for free cooling over a wider operating range, reducing reliance on compressor-based cooling. The US Department of Energy identifies direct liquid cooling as an approach with potential to improve facility energy performance.
However, direct-to-chip cooling does not necessarily remove all the heat generated within a rack. Components outside the liquid circuit may still require air cooling, making coordinated management of both systems important.
The most suitable architecture therefore depends on the IT configuration and the proportion of heat captured by liquid. Chillers, CDUs and air cooling equipment need to work together. Their combined performance determines how effectively the facility can support high-density computing while controlling the energy required to keep it operating.
This need for coordination is central to the patented cooling concept jointly developed by Daikin Applied Europe and CTS Group. The concept combines enhanced facility water systems, advanced controls, extensive free cooling and dedicated buffer pipe technology to address the operating requirements of modern AI infrastructure.
The approach considers how cooling is generated, transported and controlled as a complete system. Its buffer pipe technology helps maintain stable water temperatures supplied to liquid-cooled servers during sudden load changes and short electrical interruptions. Alongside this temperature management, the architecture is designed to maximise opportunities for efficient cooling operation.
Projected PUE values reach as low as 1.09 in Nordic climates and below 1.20 in demanding Middle Eastern environments. These are project-dependent projections, with performance influenced by climate and the proportion of IT load served by liquid cooling. They describe the overall concept rather than the isolated contribution of an individual component.
The economic relevance follows directly from the PUE calculation. For the same IT energy consumption, lowering the energy overhead reduces the electricity needed to operate the facility. Where grid capacity is constrained, a more efficient supporting infrastructure may also allow a larger share of available power to serve IT, subject to the site’s design and redundancy requirements.
The partnership combines complementary expertise. Headquartered in Switzerland, CTS Group is a global design-and-build contractor focused exclusively on data centers. Its combination of design, engineering and in-house production supports modular, prefabricated delivery, with digital tools used from initial concept through commissioning. This brings project delivery expertise alongside Daikin’s cooling equipment and application engineering. This integrated delivery model helps customers develop scalable capacity with greater programme certainty and consistent quality throughout project execution.
Daikin’s new CDU provides the interface between facility and technology cooling circuits, transferring heat while supporting the temperature and pressure requirements of liquid-cooled IT. Considering this interface alongside the chiller plant and controls helps align equipment selection with the intended operating conditions.
For operators, the value lies in evaluating these elements together from the design stage. Learn more about the Daikin–CTS cooling concept and how to select a CDU for data center liquid cooling.
Delivering the intended performance also requires a supplier able to support project execution. In the Uptime Institute Data Center Staffing and Recruitment Survey 2025, 42% of respondents to the relevant question identified manufacturer lead times among their organisation’s greatest challenges, while 17% cited inconsistent manufacturing quality.
This makes engineering support, manufacturing capability and testing experience important selection criteria. Daikin Applied Europe has a dedicated data center organisation, supported by investments exceeding €30 million in R&D and manufacturing capabilities over the past decade. Revenue grew from €442 million in FY2023 to €472 million in FY2024 and €504 million in FY2025, an increase of approximately 14% between FY2023 and FY2025.
Experience with large orders is demonstrated by OSL-Hamar in Norway, where Daikin supplied 71 free-cooling chillers and 330 fan array units. Factory Acceptance Testing capabilities up to 10 MW also enable performance verification before deployment, helping customers address technical requirements before equipment reaches site.
Daikin’s portfolio covers free-cooling chillers, CDUs, fan wall units, CRAHs and controls. This breadth supports equipment coordination across the cooling chain, while engineered skid solutions can reduce the integration work required on site.
Prefabrication brings the delivery discussion into practical focus: assembling and configuring equipment before installation can simplify site activities and commissioning. CTS Group prefabricates skids specific to the chillers and CDUs, so each package arrives ready to be integrated quickly into the system on site. Early coordination of interfaces, access and testing requirements helps project teams manage complexity and align cooling infrastructure with the construction programme.
Talk to our experts and discover the ideal Daikin cooling solution for your AI data center.