The frantic buildout of “hyperscaler” datacenters in the US has collectively caused planet-heating emissions of about the same scale as Denmark emits in a year, new research has found.
asserted
research → cause → year
In that period 403 “hyperscalers” – the largest datacenters used primarily to handle the growth of artificial intelligence – emitted about 25.7m metric tons of carbon dioxide, according to the study.
uncertain
hyperscalers → use → study
The latest research, published in Nature Sustainability, found that these datacenters were responsible for about 1.8% of total US electricity consumption.
asserted
datacenters → publish → consumption
Researchers said this vast datacenter energy demand, centered on states such as Virginia, Oregon, Ohio and Iowa, is almost certainly a conservative estimate.
asserted
demand → say → Virginia
The study used federal government data to determine emissions from the facilities over a year until April 2025.
asserted
study → use → April
Since then, the rapid expansion of datacenters has only accelerated further.
asserted
expansion → accelerate → datacenters
“There has been an explosion in these datacenters and the CO2 emissions are pretty substantial,” said Francesca Dominici, a professor of biostatistics at Harvard University and co-author of the study, which was led by Gianluca Guidi.
asserted
which → say → Guidi
“Since 2025 there have been many, many more datacenters so we need to think of this level of emissions as a lower bound.
asserted
we → need → bound
This is something that needs study and for policymakers to think about how to do this in a responsible way.”
asserted
policymakers → need → way
Datacenters use an enormous amount of energy – the new study estimates the hyperscale group consumed 68 to 99 terawatt-hours of electricity over the year – and largely draws upon dirtier sources of power to meet this demand.
asserted
group → use → demand
More than half of the generation used by hyperscalers came from the fossil fuels that are dangerously overheating our planet, the research found.
asserted
research → use → planet
The growth of AI is being underpinned by a rise in new gas-fired power, sometimes built at the site of the facilities themselves, which is being favored over installed renewables such as solar and wind.
asserted
which → underpin → solar
“Sadly, it doesn’t surprise me that the majority of hyperscale datacenters rely on fossil fuels for their power generation despite the profound and lasting climate and health impacts that could be, and should be, avoided,” said Gina McCarthy, a former climate adviser to Joe Biden’s administration who said that the study is an “important step” to assessing the impact of datacenters.
uncertain
study → surprise → datacenters
The rise of datacenters has caused alarm among climate groups amid worsening global heating that may surpass previously agreed “safe” levels and cause an acceleration of the sort of deadly events seen this year, such roasting heatwaves in Europe, wildfire smoke in North America and the disastrous landslide on the Nepal-Tibet border.
uncertain
that → cause → border
In the US, the epicenter of this AI boom, developers have defended the use of energy and water used by the facilities as marginal when compared with other industries, such as agriculture, and have touted associated jobs and raised local taxes as beneficial for communities.
asserted
developers → defend → communities
Donald Trump, a vocal supporter of the AI industry, has said people who oppose the growth of datacenters will end up “backwards and poor”.
asserted
who → say → datacenters
But a bipartisan public backlash to datacenters shows little sign of abating in the US.
asserted
backlash → show → US
Local opposition has focused on several big developments, such as a cluster of datacenters in the greater Memphis, Tennessee, region that are central to Elon Musk’s AI ambitions but have faced protests and lawsuits over their noise and air pollution.
asserted
that → focus → pollution
Musk’s AI company has responded to this by buying the homes of those who live nearby.
asserted
who → respond → those
The vast water use of one of the world’s largest planned datacenters in drought-hit Utah, meanwhile, has mobilized grassroots opprobrium.
asserted
use → hit → opprobrium
The site, called Stratos, will also emit a large amount of air pollution, a separate study recently found.
asserted
study → call → pollution
Stratos could cause health-related economic costs of up to $690m each year if it uses the dirtiest of the fossil fuel options to power itself, this research concluded.
uncertain
research → cause → itself
Burning this fuel will release up to 3,140 tons of soot-like particulate matter, linked to an array of cardiovascular and respiratory problems in people, in plumes that will waft across Utah and into Idaho, Wyoming, Colorado and Illinois.
asserted
that → burn → Idaho
“Stratos is a modern-day Trojan horse that would unleash deadly air pollution linked to serious, life-threatening health damages upon Utah communities and millions of people living downwind of this massive datacenter,” said John Walke, director of federal clean air at the Natural Resources Defense Council, which commissioned the report.
asserted
which → unleash → report
“Pollution from Stratos will contaminate the air and compromise the ability to live a healthy life both around the facility and in states hundreds of miles from the datacenter.”
The developers for the Stratos project have been planning to use combined-cycle natural gas but are also reportedly considering using renewables and other options; they have been under political pressure to reduce its scale.
uncertain
they → contaminate → scale
Last month the project filed a new water application that reportedly claimed its water demands would be lower than historical agricultural uses at the property, which some environmental critics have questioned.
uncertain
critics → file → which