New research published in Scientific Reports shows that rising global temperatures and humidity are steadily eroding the viability of direct air free cooling - the cheapest, most energy-efficient method for keeping data centers from overheating. The study, led by atmospheric scientists at the University of Hawaiʻi at Mānoa, found that the number of hours exceeding safe operating thresholds for this cooling approach has climbed significantly over the past 45 years, with the sharpest constraints emerging across the tropics and the southeastern United States.
Christina Karamperidou, atmospheric sciences professor in the UH Mānoa School of Ocean and Earth Science and Technology, led the research team. They combined high-resolution hourly weather observations with climate model simulations and a global database of data center locations. This allowed them to assess how often environmental conditions breached recommended limits for direct air free cooling - both historically and under future climate scenarios.
The timing is difficult. The AI boom is pushing data center demand higher, while the climate conditions that enable low-cost cooling are becoming less reliable. "We found that periods of time when temperature and humidity exceed recommended operating thresholds for direct air free cooling are becoming more frequent and lasting longer in many regions," Karamperidou said. "This will reduce the availability of air free cooling for a growing number of data centers globally."
How researchers measured the cooling constraint
Direct air free cooling works by pulling naturally cold outside air into a facility. It is simple, cheap, and far less energy-intensive than mechanical cooling systems. But it only works within specific temperature and humidity ranges. The research team mapped those thresholds against decades of weather data and climate projections to see where - and for how long - the method remains viable.
By analyzing global datasets, the researchers identified large-scale patterns that would not surface in studies focused on individual facilities. The share of data centers exposed to conditions that limit air free cooling availability for at least one quarter of the year is rising. Projections through mid-century point to continued expansion of these constraints as warming and humidity increase.
For those deploying AI for Science & Research workloads, the findings signal that data center siting and cooling strategy will become increasingly consequential decisions. A facility's location - once chosen for cheap power or proximity to users - now carries a climate risk component that shifts over the lifetime of the infrastructure.
Worst-day conditions are intensifying fastest
One of the study's sharper findings involves the difference between average conditions and extremes. In several regions, the most stressful days are worsening faster than the mean. That gap matters for anyone responsible for keeping a data center running.
"From an operational perspective, those worst-day conditions often drive contingency planning, system overrides, redundancy requirements, and reliability decisions," Karamperidou said. "This suggests that infrastructure planning may need to account not only for average environmental conditions but also for how the most stressful days are changing over time."
The research was developed with collaborators working on cyberinfrastructure and next-generation cooling technologies, including partners through the National Science Foundation ERC EARTH, where UH is a key partner institution.
Why this matters for science and research professionals
Research computing depends on reliable, affordable data center capacity. When cooling costs rise or availability drops, budgets feel it - and so do project timelines. This study provides a quantitative basis for factoring climate projections into infrastructure decisions that have traditionally relied on historical weather norms. For researchers and lab managers planning long-term compute resources, the message is straightforward: the cooling assumptions that worked for the past two decades will not hold for the next two. Site selection, backup cooling capacity, and worst-day contingency plans all need recalibration against forward-looking climate data, not rear-view averages.
Your membership also unlocks:
