ENIAC, the world’s first programmable electronic general-purpose digital computer, was built at the University of Pennsylvania in 1945. Originally designed to calculate complex artillery firing tables for the U.S. Army, ENIAC’s first major program was to study the feasibility of the H-bomb. Nicknamed “the Giant Brain,” its speed was dazzling: a calculation that previously took 12 hours could be completed in just 30 seconds. Over the next 11 years, ENIAC took on more complex tasks, growing in capacity and physical footprint.  By the time the Giant Brain retired in 1956, it occupied around 300 square feet and was supplied with 150 kW of electricity at a time when the standard supply for an average American home was 60 amps (7.2kW). 

I don’t imagine anyone in 1956 worried much about the electrical demand or physical footprint of ENIAC. But ENIAC’s technological progeny—the massive data centers using generative artificial intelligence (AI)—operate on a scale of energy consumption previously undreamed of. Facilities are measured in millions of square feet and terawatt hours (TWh) of energy consumption. Environmental advocates, public policy experts and local and state governments have voiced alarm about the potential impacts, including emissions associated with increased energy generation, water consumption, noise, and the loss of agricultural lands and local autonomy in zoning decisions.   

Now researchers have identified an additional cause for concern. At the end of March, an international team from the U.K., Singapore, France, Italy, Genoa and Hong Kong released a preliminary version of a new study titled “The Data Heat Island Effect: Quantifying the Impact of AI Data Centers in a Warming World.”The team looked at AI hyperscalers around the world, giant facilities (as much as a million square feet) that house thousands of servers. They discovered significant increases in land surface temperature (LST) in the surrounding areas, ranging from 2 degrees Fahrenheit to as much as 16 degrees Fahrenheit.  They concluded that heat dissipation, directly connected to the energy consumption for computing and the cooling technology required for the operation of AI hyperscalers, is creating heat islands, warmer microclimates that can affect areas as much as 6.2 miles away, with the potential to affect 340 million people around the world who live in proximity to these facilities.

To isolate temperature changes specific to large-scale data centers, the study focused on 8,472 facilities located outside urban centers. The analysis used data on land surface temperature (LST) measurements recorded by NASA’s MODIS (Moderate Resolution Imaging Spectroradiometer) instruments, using a dataset produced by NASA acquired between 2004 and 2024. A remarkable device that NASA has compared to a Swiss army knife, MODIS is a key instrument aboard NASA’s Terra and Aqua satellites, which have viewed the Earth’s entire surface every one to two days for the past 26 years. In addition to LST, MODIS collects all kinds of invaluable data, including changes to multiple types of vegetation (such as forests, grasslands and croplands), bare soils and water, cloud characteristics, moisture profiles, snow cover, aerosols and water vapor. 

The study, which still awaits peer review, notes that the observed effects of hyperscaler facilities are similar to the very well-documented effects of urban heat islands, in which factors such as a pavement, heat-absorbing building materials, vehicles, air conditioning equipment and industrial activities, coupled with the reduced presence of forested land, water bodies and green space, create pockets of heat that can raise daytime temperatures by 1 to 7 degrees Fahrenheit, and nighttime temperatures by 2 to 5 degrees Fahrenheit.  Heat islands cause higher levels of carbon dioxide through the need to generate more electricity with fossil fuels to meet cooling needs, which in turn contributes to the formation of fine particulate matter and ground-level ozone. Heat combined with these increased pollutants can cause health problems and even death for older adults, young children, people who work outdoors, and people with chronic health conditions, disabilities, mobility constraints and those taking certain medications.

The researchers conclude that the hyperscale data centers are creating warmer local climate zones, which they call the “data heat island effect.” The effect is significant. It has implications for health, energy, demographic distribution, wildlife, global warming and climate transformation. Since the expectation is that these facilities will continue to proliferate in the foreseeable future, the study recommends that data heat islands should be an important factor in planning and decision-making. 

The paper is not all doom and gloom. It proposes both software and hardware solutions for better energy efficiency, as well as system engineering and even passive building design strategies to improve cooling while lowering heat dissipation. Most energy efficiency measures are a win-win proposition. Industry insiders who have already begun attacking the study might want to take a closer look.