AI data centres are heating the planet and millions may feel it

Artificial intelligence’s explosive growth is leaving a measurable footprint on the physical world, with new research showing that data centres are not just consuming vast amounts of electricity but also warming the land around them.

A study by researchers from institutions including the University of Cambridge finds that areas surrounding AI data centres experience an average land surface temperature increase of about 2 degree Celsius after operations begin, creating what the authors describe as a “data heat island effect.” 

“The strong and continuous increase of AI-based services leads to the steady proliferation of AI data centres worldwide with the unavoidable escalation of their power consumption,” the authors wrote. 

New kind of heat island

The phenomenon mirrors the well-known urban heat island effect, where cities trap heat due to dense infrastructure, but is now being driven by compute.

Using two decades of satellite-derived land surface temperature data, the researchers identified a “consistent gradient in temperature between the area of data centres and their surrounding regions,” adding that the effect is large enough to “form a ‘data heat island effect.’” 

The findings show a sharp temperature jump coinciding with the start of data centre operations. Average increases were measured at 2.07 degree Celsius, with extremes reaching as high as 9.1 degree Celsius in some locations. 

This puts AI infrastructure in the same league as traditional industrial heat drivers. Urban heat islands typically raise temperatures by 4-6 degree Celsius, suggesting data centres could become a comparable localised climate force.

Heat travels further than expected

The warming effect does not stay confined to server farms. The study finds that elevated temperatures extend up to 10 kilometres from data centre sites, with a 1 degree Celsius rise measurable as far as 4.5 kilometres away. 

“The impact of data centres and AI hyperscalers activities on climate might not be negligible,” the researchers warned, noting that energy-intensive AI workloads amplify both heat dissipation and emissions. 

Case studies in regions such as Spain’s Aragón, Mexico’s Bajío, and Brazil’s northeast show anomalous temperature increases aligning with data centre clustering, reinforcing the link between compute infrastructure and localised warming.

Growing human impact

More than 340 million people globally could already be exposed to these localised temperature increases, the study estimates, even though many facilities are built outside dense urban zones. 

“This makes the data heat island effect a phenomenon that is very hard to consider negligible,” the authors said, warning of potential knock-on effects on “welfare, healthcare and energy systems.” 

AI’s energy problem is becoming physical

Data centres are expected to become among the most power-hungry sectors in the coming decade, with consumption for computation potentially exceeding manufacturing energy budgets within years. 

Compounding the issue, most AI infrastructure still relies heavily on fossil fuels, meaning rising compute demand directly translates into higher emissions and heat output.

Yet quantifying AI’s environmental cost remains difficult. “There are no credible indicators that can help determine and quantify the impact of existing AI applications,” the researchers noted, citing gaps in data and inconsistent methodologies.