Artificial intelligence data centers generate enough heat to raise temperatures in surrounding areas, according to new research examining how these facilities influence land surface temperatures over time.
Scientists found that land near AI data centers can warm by an average of 2°C (3.6°F) within months after operations begin, with extreme increases reaching 9.1°C (16.4°F). The warming effect can extend as far as 10 kilometers (6.2 miles) from the facilities and may already affect more than 340 million people who live within that distance.
The research was led by Andrea Marinoni of the Earth Observation group at the University of Cambridge in the United Kingdom. Marinoni and colleagues began investigating the environmental effects of data centers after observing that the energy required to operate these facilities has been increasing and is expected to “explode” in the coming years. Data centers that support artificial intelligence depend on large computing clusters and cooling systems that require large amounts of electricity.
To study temperature changes, the team analyzed satellite measurements of land surface temperatures collected over the past 20 years. They compared those measurements with the geographic coordinates of more than 8,400 AI data centers across the world. Because surface temperature can change due to industrial activity or heating in densely populated areas, the researchers concentrated on facilities located away from highly populated regions. More than 6,000 data centers met those conditions and were included in the analysis. Seasonal changes, global warming trends, and other environmental influences were filtered out to isolate the effect of the facilities themselves.
After a data center began operating, nearby land surface temperatures increased by an average of 2°C (3.6°F). In extreme cases, the increase reached 9.1°C (16.4°F). The temperature rise was detectable beyond the facility boundaries. Measurements showed elevated temperatures as far as 10 kilometers (6.2 miles) away. At a distance of seven kilometers, the intensity of the warming decreased by only about 30 percent.
“The results we had were quite surprising,” Marinoni said. “This could become a huge problem.”
Population data used in the study indicates that more than 340 million people live within 10 kilometers of data centers. Those communities experience warmer land surface temperatures than they would if the facilities had not been built.
Regional patterns provide examples of the trend. In Mexico’s Bajío region, which has developed into a data center hub, temperatures increased by about 2°C (3.6°F) between 2004 and 2024. A similar increase of about 2°C (3.6°F) occurred in Aragon province in Spain, which hosts hyperscale AI data centers. Neighboring provinces did not record comparable increases.
The facilities involved are often called AI “hyperscalers.” These complexes contain thousands of servers and can extend across more than one million square feet. Most of them were constructed during the past decade as demand for artificial intelligence computing expanded.
Real estate firm JLL estimates global data center capacity will double between 2025 and 2030, with artificial intelligence accounting for half of that demand. BloombergNEF projects that electricity demand from data centers will more than double by 2035.
The study has not been peer-reviewed and is available as a preprint on arXiv.
Chris Preist of the University of Bristol said additional research would help determine the exact sources of the warming. “It would be worth doing follow-up research to understand to what extent it’s the heat generated from computation versus the heat generated from the building itself,” he said, adding that sunlight heating the building structure could contribute to the effect.
Ralph Hintemann, a senior researcher at the Borderstep Institute for Innovation and Sustainability, said the study provides “some interesting figures” but said the reported effects “seem very high.” He added that emissions from electricity generation used to power data centers remain the more alarming aspect in relation to climate change.
Deborah Andrews, emeritus professor of design for sustainability and circularity at London South Bank University, said the research is the first paper she has seen examining heat produced by data centers. “The ‘rush for AI-gold’ appears to be overriding good practice and systemic thinking,” she said, “and is developing far more rapidly than any broader, more sustainable systems.”
Data centers also require extensive land, water, and electricity. Technology companies including Amazon Web Services, Google, Meta and Microsoft have purchased large properties to construct artificial intelligence facilities.
Land purchases connected to these projects have produced local disputes. In Kentucky, a woman and her mother declined a $26 million offer to sell their land to an unidentified buyer described as a “major artificial intelligence company,” according to WLKY.
Energy consumption by data centers has drawn attention from lawmakers. Democratic Sen. Elizabeth Warren and Republican Sen. Josh Hawley sent a letter to the Energy Information Administration on Thursday requesting the agency to “establish a mandatory annual reporting requirement for data centers.” The senators wrote that collecting energy data would assist with grid planning and “will support policymaking to prevent large companies from increasing electricity costs for American families.”
Legislation concerning data center construction has also been proposed. Rep. Alexandria Ocasio-Cortez and Sen. Bernie Sanders introduced a bill calling for a pause on new data center construction until government safeguards are created.
“AI and robotics are creating the most sweeping technological revolution in the history of humanity,” Sanders said in a statement. “The scale, scope and speed of that change is unprecedented. Congress is way behind where it should be in understanding the nature of this revolution and its impacts.”
Marinoni said discussion about managing artificial intelligence infrastructure remains possible. “There still might be time to consider the possibility of a different path … without affecting the demand of AI and its ability to provide progress for mankind,” he said.
