As technology giants pour hundreds of billions of dollars into artificial intelligence (AI), a new study has exposed the staggering environmental cost of the technological revolution.
Researchers at Harvard University have discovered that just 403 of America’s largest data centres consume nearly as much electricity as entire US states.
The study, published in Nature Sustainability, estimated that these facilities consumed a whopping 82 terawatt-hours of electricity between May 2024 and April 2025.
That is enough electricity to power nearly seven million American households for an entire year.
Approximately 54% of the electricity powering these facilities came from fossil fuels, resulting in an estimated 25.7 million tonnes of carbon dioxide emissions.
The figure exceeds Denmark’s annual carbon dioxide emissions.
Could the situation get worse?
Speaking to CBS News, the study’s lead author, Gianluca Guidi, revealed that the number of hyperscale data centres operating in the United States has more than doubled since the study period ended.
Moreover, another 2,000 facilities are planned or under development.
These enormous facilities house thousands of computers that operate around the clock, requiring constant cooling and consuming tremendous amounts of electricity.
The findings come as negotiators at the COP31 climate summit in Turkiye launched a voluntary pledge urging technology companies to power their AI infrastructure using cleaner energy.
However, major American technology companies driving the AI revolution have yet to sign the pledge.
Meanwhile, growing electricity demand has already contributed to higher electricity bills in some American communities, raising questions over who ultimately pays for the AI revolution.
