November 20, 2026
Warmer Water, Lower Electricity: Holistic Thermal Management for More Sustainable Data Center Design
A technical perspective on how to achieve more efficient data center cooling through hybrid systems that use warmer-water cooling and mechanical chilling to meet requirements of today and tomorrow.
Data centers are under increasing pressure to support higher-density compute while reducing electricity use and improving sustainability. Recent advances focus on reducing losses in the electrical infrastructure design as well as in the thermal system design. Electrical improvements such as 800 VDC can reduce losses in power conversion and distribution. Thermal advances such as liquid cooling and higher allowable fluid temperatures can reduce the electricity required to remove heat from high-density compute systems.
After the IT infrastructure, heat rejection is one of the most energy-intensive processes in the facility. Recent advancements in chip packaging and silicon design are enabling some processors to operate reliably with higher technical cooling fluid temperatures. Used appropriately, this can reduce the electricity required for mechanical cooling, lower annual energy use and free up critical energy capacity that can be redirected to compute workloads.
Realizing the benefits of higher-temperature liquid-cooling capability in real-world facility design requires collaboration across the data center cooling ecosystem. Because those benefits depend on site conditions, climate and the overall facility design, thermal management solutions need to be flexible and responsive to changes in load and to future generations of IT equipment. To support this evolution, Trane Technologies is focused on optimized solutions that can support a range of technical fluid temperatures and chip architectures. One practical path forward is an integrated system that can respond to changing operating conditions in real time to reduce electricity use while maintaining reliable thermal performance.
Overview of thermal management for data center cooling
The purpose of a data center thermal management system is to maintain the IT system within allowable operating temperatures by continuously moving excessive heat from the chips, servers and racks to the ultimate heat sink, typically outdoor ambient air. The efficiency of this process depends on thermal resistance, the energy required at each step of the chain and the overall system design. Higher cooling fluid temperatures do not reduce the amount of heat the facility must reject; they change how much energy may be required to move that heat under different operating conditions.
Warmer technical fluid temperatures can increase the number of hours when free cooling is possible, because outdoor conditions in many locations are often cool enough to support non-mechanical heat rejection. However, even when higher chip temperature limits improve annual efficiency and reduce Power Usage Effectiveness (PUE), most data centers still require mechanical cooling at some point during the year to maintain the IT system within allowable temperature limits.
The sections below examine the thermodynamic and operational realities that shape data center cooling design, including how higher fluid temperatures can reduce electricity use for cooling in some conditions and where mechanical cooling may still play a role as part of a hybrid integrated system.
Design considerations for warm water cooling
The maximum allowed temperature to the rack is just that—a cap or “shall not exceed” value. Technology Cooling System (TCS) supply temperature setpoints need to be less than the maximum value to provide a safety margin for load transients. This margin is especially important for AI workloads, where both training and inference environments can create rapidly changing thermal demand due to workload variability. When compute loads spike, Coolant Distribution Units (CDUs) and secondary loops require time to respond to the increased thermal load. To maintain an adequate safety factor during these swings, most operators design around cooler CDU supply-water temperatures — for example, operating around 41°C (105.8°F) instead of continuously operating at an upper limit such as 45°C (113°F).
The thermal resistance chain: why one temperature limit does not define system design
The thermal management system’s objective is to keep the IT equipment within its thermal design envelope, transporting excessive heat away from it. The design of such a system must account for all fluid flows and temperatures, as well as thermal resistances that exist between the chip and the ultimate heat sink.
Heat moves through a series of interfaces: from the chip to the cold plate, through the technical loop, across the CDU, into the facility loop and finally to the outdoor environment or a heat reuse application (Figure 1). At each of these steps, the heat transfer process is non-ideal, which results in a stack-up of temperature differences to make heat flow. Under hotter ambient conditions, peak loads or degraded real-world heat exchanger performance, mechanical cooling can reject the required amount of heat while avoiding excessive chip temperatures. This is why chip or rack temperature limits alone do not define facility cooling requirements, and why thermal buffer and operating margin should be built into the thermal management system.
The data center microclimate: heat islands, variability and real-world conditions
In addition to the thermal resistance challenge, localized microclimate or heat island effects further influence the thermal management system and the ability to utilize free cooling. When calculating the viability of free cooling, standard models often rely on regional hourly weather data, which represents broad climatic conditions but does not capture the localized thermal effects created by high-density heat rejection equipment. Gigawatt-scale heat rejection creates a complex interaction between the facility’s heat rejection infrastructure and the ambient air surrounding the cooling plant.
A hyperscale data center using large arrays of dry coolers can create a very localized artificial heat island, or microclimate, over the heat rejection systems (such as dry coolers), raising the effective ambient dry-bulb temperature seen by parts of the cooling plant by as much as 5°C (9°F), and in extreme situations, up to 20°C (35°F) (Figure 2). If those effects are not included in free cooling assessments, regional weather data alone can overstate available cooling capacity during peak conditions.
When these location and equipment specific microclimate effects are considered alongside long-term peak ambient data, the practical limits of ambient-only heat rejection become more apparent. While warm technical cooling fluid can substantially increase the number of hours when heat can be rejected without mechanical cooling, it does not guarantee 100% free cooling across the United States. For many data center locations, some mechanical cooling is still required on hot days. Analysis of representative European locations shows a similar result: more free cooling, but not 100% year-round operation in most locations (Figure 3).
Optimizing the mechanical plant
Higher cooling fluid temperatures can increase the number of hours per year that free cooling, or economization, can be used, reducing the electricity used for cooling. But free cooling is not without its own energy and infrastructure costs.
Relying entirely on dry coolers for free cooling requires large fan arrays moving very high volumes of air. This introduces parasitic fan power, increases equipment footprint and can create substantial acoustic challenges. Adiabatic assist and other indirect evaporative strategies can extend free cooling performance, but they do so at the expense of water use.
In high ambient temperature conditions, the quantity and footprint of dry coolers needed to reject data center heat can become economically and spatially impractical. Integrated systems that combine smart chillers with free cooling can provide greater flexibility across a wider range of outdoor conditions.
At low ambient temperature conditions, free cooling can reject heat with relatively low electricity use. At high ambient temperature conditions, chillers may be needed to provide the incremental cooling required to maintain capacity and control. Between those extremes, mixed-mode operation (combining partial mechanical cooling with free cooling) can reduce cooling energy use.
Data center operators must therefore balance the energy required for mechanical cooling against the fan power, footprint, acoustics, and - where applicable - water use associated with maximizing convective free cooling.
Planning for higher densities and future uncertainty
The pace of change in AI infrastructure requires flexibility. Chip manufacturers are publishing roadmaps that show significant shifts in power density and thermal design from one generation to the next. In contrast, data center facilities are typically designed for service lives of 15 to 20 years. This creates a basic design challenge: infrastructure built today must support future thermal requirements that are not yet fully known.
Future computing advances may change cooling requirements through improvements in chip packaging, near-chip hotspot management or alternative thermal interface or cold plate materials. But it remains uncertain whether those advances will continue to keep pace with rapidly rising chip power levels, or whether future platforms will operate effectively using warmer cooling fluid temperatures.
One way to mitigate this uncertainty is optimally integrating chillers in the thermal management system. Adopting solutions that combine mechanical and free cooling can help data center owners support a wider range of supply temperatures and adapt to future computing architectures without being locked into a single operating envelope.
The future is flexible
Based on all the above factors, we believe that data center designs should choose an intelligent, hybrid architecture that optimizes for token output, annual and peak energy use, acoustics, time-to-token and total cost of ownership.
Higher TCS fluid temperatures can increase free cooling hours and reduce annual energy use. But thermal resistance, microclimate effects, peak ambient conditions, mixed compute loads and long-term infrastructure uncertainty reinforce the need for a hybrid thermal management system that includes dry coolers and mechanical cooling in most applications.
For the foreseeable future, the most efficient facilities will be those that use climate-optimized intelligent hybrid heat rejection systems that maximize free cooling when conditions allow, while preserving mechanical cooling capacity for reliability, flexibility and year-round operation.
Engineering for today and tomorrow
The transition toward warmer technical cooling loops is an important development in data center thermal management. It opens the door to hybrid heat rejection strategies that shift dynamically between free cooling and mechanical cooling as outdoor conditions change, delivering significant efficiency gains and helping preserve more electrical capacity for compute. This also leads to more sustainable solutions that minimize the use of electrical energy, water and noise emission while enabling the operation of high-performance data centers.
At Trane Technologies, we believe a holistic systems-based approach to thermal management will help data center owners balance competing design requirements and adopt sustainable solutions. Maximizing efficiency requires balancing chip performance, water temperature, fan power, site conditions and mechanical cooling as part of one integrated system. We are committed to helping customers and ecosystem partners design robust, climate-optimized facilities that can meet the growing demands of AI more sustainably and with lower electricity use.
Topic Tags
English