The Heat Behind Artificial Intelligence: Are Data Centres Creating a New Environmental Footprint?
As artificial intelligence drives an unprecedented expansion of digital infrastructure, the environmental conversation has concentrated on electricity, water and carbon emissions. But there is another question that deserves attention: what happens to the enormous quantities of heat these facilities continuously release into the surrounding environment?
The expansion of artificial intelligence is creating a new industrial economy, one built not around factories, assembly lines and traditional manufacturing, but around processors, high-density computing systems and enormous data-centre campuses. Governments are competing to attract these investments, technology companies are committing billions of dollars to infrastructure, and electricity providers are preparing for demand that would have been difficult to imagine only a decade ago. Much of the public discussion has focused on whether the electricity supply can keep pace, how much water cooling systems consume and what the expansion means for carbon emissions.
There is, however, another environmental consideration that deserves considerably more attention. Every data centre generates heat, and that heat must be transferred somewhere. Cooling systems protect the equipment by removing thermal energy from the buildings, but they do not destroy it. They ultimately release it into the surrounding environment through cooling towers, condensers, radiators, heat exchangers and ventilation systems. As the number, size and concentration of these facilities increase, we should be asking whether the cumulative thermal output is creating a new environmental footprint around the infrastructure powering the digital economy.
This is not an argument against artificial intelligence or the development of data centres. It is an argument for understanding the complete environmental consequences of infrastructure before its expansion outpaces the ability of communities and regulators to manage them.
The energy does not disappear. It becomes heat.
To understand the issue, it helps to begin with a basic principle of thermodynamics. Electricity consumed by computing equipment is ultimately converted into heat. Processors, memory systems, networking equipment, power supplies, lighting and other electrical infrastructure all contribute to the facility’s thermal load. Almost all the electricity entering a data centre eventually has to be accounted for as heat, although the precise location and method of heat rejection depend on the facility’s architecture.
Consider a hypothetical data centre drawing 100 megawatts of electricity continuously. Its electrical consumption represents approximately 100 megawatts of energy being converted into heat every second. Over 24 hours, that amounts to 8,640 gigajoules of energy. Over an entire year, the figure reaches approximately 31.5 petajoules.
Scale that infrastructure to 500 megawatts and the daily thermal equivalent becomes 43,200 gigajoules. A one-gigawatt facility operating continuously represents approximately 86,400 gigajoules of energy each day. These figures are simplified energy-balance calculations. They assume continuous operation and that essentially all electrical energy ultimately becomes heat requiring rejection. They are not measurements of the temperature increase in the surrounding atmosphere. That distinction matters because the amount of energy released and the temperature change experienced by a neighbouring community are two different measurements.
Nevertheless, the calculations illustrate the scale of the physical process. A hyperscale data-centre campus is not simply a large building containing computers. It is a continuously operating thermal system, transferring enormous quantities of energy from its interior into its surroundings.
The cooling system changes the destination, not the underlying problem
The technology used to remove heat is important because different cooling architectures produce different environmental consequences. Traditional air-cooling systems move air through server racks and transfer the resulting heat to mechanical cooling equipment. Liquid-cooling systems circulate coolant through components or heat exchangers, allowing thermal energy to be captured more efficiently, particularly in high-density computing environments. Evaporative systems use the phase change of water to remove heat, while dry cooling systems transfer heat to outside air through heat exchangers without relying on evaporation.
Hybrid arrangements combine several of these technologies, depending on climate, water availability, electricity costs and the facility’s operating requirements. The important distinction is that these systems manage the movement of heat rather than eliminate it. A facility may substantially reduce its water consumption by adopting dry cooling, but it still has to reject thermal energy. Another facility may use evaporative cooling to reduce mechanical electricity consumption, but that approach introduces water consumption and transfers heat through the evaporation process.
This creates an environmental trade-off that cannot be adequately understood by looking at a single efficiency metric. A data centre might perform well in terms of electricity efficiency while placing additional pressure on local water resources. Another might reduce water consumption while increasing the electrical demand associated with mechanical cooling.
The American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE)
now identifies liquid cooling, dry coolers, climate-appropriate economisation and heat recovery as important considerations for AI data-centre efficiency. Its guidance also recognises that higher-temperature liquid-cooling systems can make waste heat more suitable for productive reuse. (ASHRAE)
The question, therefore, is not simply whether a cooling system works. It is whether the entire system is environmentally appropriate for the location in which it operates.
A new category of local environmental impact is emerging
During the industrial age, communities became familiar with the environmental consequences of factories, refineries, power stations, steel mills and other large industrial facilities. Smoke, particulates, chemical emissions, wastewater and industrial noise eventually became subjects of regulation, monitoring and public accountability. Thermal pollution also has a long history. Power stations have discharged heated water into rivers and other bodies of water, sometimes altering aquatic environments and affecting the organisms living there. Industrial facilities have released waste heat into the atmosphere for generations.
Data centres are not the first infrastructure to generate substantial waste heat. What distinguishes the present expansion is the concentration of electrical computing capacity, the continuous nature of the operations and the possibility of multiple large facilities being clustered within relatively small geographical areas. Unlike many traditional manufacturing operations, data centres can maintain substantial thermal output throughout the day and night. Their cooling infrastructure may operate continuously, including during periods when surrounding residential communities would ordinarily experience a decline in daytime heat.
This raises questions about the relationship between data-centre development and the urban heat island effect. Urban heat islands occur when built environments become warmer than surrounding areas because of materials such as concrete and asphalt, reduced vegetation, building density and changes to airflow. Introducing another concentrated source of anthropogenic heat into such an environment creates an additional variable that should be measured rather than assumed away.
The important question is whether the heat rejected by data centres produces measurable temperature changes beyond the facility boundary, how far those changes extend, and under what atmospheric conditions they become significant.
Phoenix provides an important real-world warning
In May 2026, researchers David J. Sailor, Soroush Samareh Abolhassani and Eli P. Martin published a peer-reviewed study in the Journal of Engineering for Sustainable Buildings and Cities examining the thermal effects of operational data centres in the Phoenix metropolitan area. The research is particularly important because it moved beyond theoretical modelling and examined actual outdoor temperatures around four facilities. The researchers conducted vehicle-based temperature measurements in residential neighbourhoods located downwind of the data centres.
Their findings identified average downwind air temperatures approximately 0.7 to 0.9 degrees Celsius higher than corresponding upwind measurements. The maximum observed temperature difference reached approximately 2.2 degrees Celsius, with detectable thermal signatures extending as far as 500 metres from facility boundaries.
The facilities examined included a 36-megawatt data centre in Mesa and a 169-megawatt campus in Chandler. The researchers estimated that the larger campus rejected an amount of waste heat equivalent to the electricity consumption of more than 180,000 households, concentrated within a relatively small geographical footprint. (Arizona Board of Regents)
These findings deserve attention, particularly because Phoenix already experiences significant heat exposure. However, scientific accuracy requires a distinction: the study examined specific facilities, locations, wind conditions and measurement periods. It does not establish that every data centre produces the same temperature increase or that all surrounding neighbourhoods will experience identical effects. It does establish something important. Under certain conditions, the thermal effects of data-centre cooling infrastructure can be measured outside the facility itself. That should be sufficient reason for planning authorities to begin treating thermal discharge as a legitimate component of environmental assessment.
New research from Japan complicates the picture
A second study, published on November 4, 2026, provides an important counterpoint.
In Satellite Observations and Energy Balance Constrain the Thermal Footprint of a Hyperscale Data-Centre Cluster, researcher Jiawei Jing examined satellite observations covering 13 years of development around data-centre clusters in Inzai and Tama, Japan.
The research followed the development of 12 facilities between 2011 and 2025. Rather than relying exclusively on temperature measurements near cooling equipment, it examined satellite-derived land-surface temperatures and compared changes during construction and after facilities became operational.
The findings challenge the assumption that large data-centre clusters necessarily create widespread neighbourhood-scale ground warming. The study reported that daytime land-surface temperatures over newly developed facilities fell by approximately 1.8 to 2.2 degrees Celsius, associated with the reflective characteristics of engineered roofs replacing previous land surfaces. At night, warming across the wider clusters remained below 0.3 degrees Celsius. (DOI)
This is an important qualification, not a reason to dismiss the Phoenix findings.
The two studies examine different physical phenomena. The Phoenix research measured near-ground air temperatures along routes downwind of cooling infrastructure. The Japanese research examined satellite-derived land-surface temperatures across much larger geographical areas and over an extended period.
Air temperature and land-surface temperature are not interchangeable. A plume of warm air can produce measurable temperature differences near a facility without creating a comparable warming signature across an entire satellite-observed neighbourhood. Similarly, changes in roofing materials, vegetation and land use can influence satellite surface-temperature readings independently of the heat rejected by operating servers.
The Japanese research also found that much of the nighttime surface-temperature change occurred during construction, before the facilities were fully operational. That finding illustrates why environmental studies must distinguish between the effects of the buildings themselves, the conversion of land and the thermal output of the computing equipment.
Taken together, these studies offer a more useful conclusion than either extreme. Data-centre heat can produce measurable local atmospheric effects, but the existence, magnitude and geographical extent of those effects depend on the facility, the surrounding environment and the measurement being used. The scientific task is now to establish where those effects occur, how long they persist and which communities are exposed.
The scale of expansion makes the question more urgent
The International Energy Agency’s Energy and AI report estimated that data centres consumed approximately 415 terawatt-hours of electricity globally in 2024, equivalent to roughly 1.5 per cent of global electricity consumption. Its base-case projection placed consumption at approximately 945 terawatt-hours by 2030, more than doubling the 2024 figure. (IEA Blob Storage)
That projection is not a forecast of temperature increases. It is a measure of the electrical energy expected to pass through data-centre infrastructure, and therefore an indication of the scale of the thermal-management challenge. If the projected electricity consumption materialises, the average electrical load represented would be approximately 108 gigawatts worldwide. Since nearly all the electricity consumed ultimately becomes heat, the infrastructure would be managing an enormous continuous thermal flow.
However, the direct heat released by data centres should not be confused with greenhouse-gas emissions or treated as an equivalent mechanism of global warming. Waste heat disperses and eventually radiates into space. Greenhouse gases alter the planet’s radiative balance by affecting how energy is retained in the atmosphere. The principal global climate concern remains the emissions associated with electricity generation and the wider supply chain.
The direct release of waste heat is a separate issue, with its most immediate consequences likely to be local and regional rather than a direct driver of global climate change. That distinction is essential. We should neither exaggerate the global warming contribution of data-centre heat nor overlook its potential effects on the communities surrounding individual facilities.
The question of environmental justice
The geographical distribution of data-centre infrastructure deserves closer examination.
A company may identify a location based on land availability, access to electricity, fibre-optic connectivity, tax incentives and proximity to other infrastructure. Those are commercially understandable considerations. But the suitability of a location should also be evaluated from the perspective of the people who already live there.
A community may be asked to accommodate the physical consequences of a facility while receiving only a fraction of its economic benefits. Residents may experience additional noise, construction disruption, electricity infrastructure expansion, water demand and potentially localised thermal effects. The jobs created by highly automated facilities may not necessarily be proportional to the scale of the investment or the resources consumed.
This is not an assertion that every data-centre development creates environmental injustice. It is a question of whether the distribution of benefits and burdens is being properly assessed.
Research into the environmental footprint of data centres has already highlighted the importance of location. A 2021 study by researchers associated with Lawrence Berkeley National Laboratory examined the geographical distribution of data-centre energy, water and carbon footprints in the United States. It found that a substantial share of direct data-centre water use was associated with servers located in moderately to highly water-stressed watersheds. (Energy Analysis Division)
Thermal exposure adds another dimension to that discussion. The environmental assessment of a proposed facility should not stop at the property boundary if its cooling infrastructure can influence the surrounding atmosphere.
We should stop treating waste heat as an unavoidable afterthought
There is an opportunity to approach this problem differently. Instead of designing data centres solely around the removal of heat, developers could increasingly design them around the recovery and productive use of that heat. The principle is straightforward. If a facility continuously produces thermal energy, some of that energy may be useful to another nearby system. Potential applications include district heating, commercial buildings, domestic hot-water systems, greenhouses, agricultural operations and certain industrial processes. In colder climates, a data centre could potentially supply heat to neighbouring buildings rather than simply rejecting all of it into the atmosphere.
Research published in Energy and Buildings in 2024 by researchers at Oak Ridge National Laboratory examined the use of high-temperature heat pumps to upgrade low-grade data-centre waste heat for district-heating applications. The study evaluated configurations capable of supplying heat at temperatures of up to 120 degrees Celsius. (Oak Ridge National Lab)
ASHRAE’s more recent AI data-centre guidance also encourages operators to consider warm-water cooling loops, district-heating connections and energy-reuse metrics. It recognises that direct-to-chip liquid cooling can capture heat at temperatures more suitable for productive reuse than conventional low-temperature air exhaust. (ASHRAE)
This is an important shift in thinking. Waste heat does not have to be regarded exclusively as an operational burden. Under the right conditions, it can become a recoverable energy resource. There are practical limitations. Heat cannot be transported economically over unlimited distances, and its usefulness depends on temperature, demand, infrastructure and the availability of a suitable recipient. A data centre located far from buildings or industrial facilities that need heat may have limited opportunities for recovery.
That is precisely why these considerations belong in the planning stage. The location of a data centre should be evaluated not only in relation to electricity and fibre networks, but also in relation to its potential thermal environment and opportunities for energy recovery.
The missing environmental assessment...
The expansion of digital infrastructure is moving quickly. Environmental planning and regulatory frameworks need to move with it. Before approving large data-centre developments, planning authorities should consider requiring a thermal-impact assessment alongside existing environmental and infrastructure reviews. Such an assessment should examine the projected heat-rejection capacity of the facility, the type and location of cooling equipment, prevailing wind conditions, seasonal temperature patterns and the proximity of residential communities. It should distinguish between daytime and nighttime effects and assess the cumulative impact of multiple facilities operating within the same geographical area.
Importantly, the assessment should establish a baseline before construction and provide for monitoring after the facility becomes operational. Without baseline measurements, it becomes difficult to distinguish the thermal effects of a data centre from those of new buildings, paved surfaces, vegetation removal and other changes in land use. The monitoring should also distinguish air temperature from land-surface temperature. Satellite imagery, ground-based sensors, meteorological measurements and atmospheric-dispersion modelling each provide different information. No single method should be treated as a complete account of the environmental impact.
Developers should also disclose projected electricity consumption, cooling-water requirements, heat-rejection capacity and opportunities for waste-heat recovery. Where several facilities are planned within the same region, authorities should evaluate their combined effects rather than assessing each project as if it existed in isolation. The objective is not to create unnecessary obstacles to technological development. It is to ensure that infrastructure projects are evaluated according to their full physical and environmental footprint.
The next industrial question
Artificial intelligence is becoming one of the most consequential technological developments of this generation. Its applications are expanding across healthcare, manufacturing, finance, transportation, scientific research and public administration. The infrastructure supporting that transformation will continue to grow, and the demand for computing capacity will remain an important consideration for governments and investors.
But technological progress does not exempt infrastructure from environmental accountability.
The industrial age taught societies that the economic benefits of production could not be evaluated separately from the pollution, waste and environmental consequences left behind. It took decades of research, public pressure and regulatory development to establish the standards that now govern many traditional industrial activities. The digital infrastructure age should not have to repeat that process.
We should be asking whether the facilities powering artificial intelligence are being designed with the surrounding environment in mind, whether their waste heat can be recovered, and whether communities are being adequately protected from the cumulative consequences of their development. The evidence is still developing. The Phoenix measurements demonstrate that local atmospheric warming can occur around operational data centres. The Japanese satellite research shows that broader surface-temperature effects are more complicated and cannot simply be inferred from facility energy consumption. Together, they make the case for better measurement, not premature conclusions.
The central issue is straightforward: a data centre may be designed to keep its machines cool, but the heat removed from those machines does not disappear. It becomes part of the environment into which the facility releases it. As the world builds the physical infrastructure of artificial intelligence, environmental responsibility must extend beyond the electricity entering the building and the water used to cool it.
Comments