US admits data centers are harmful to health

The rise of artificial intelligence over the past two years has sparked a frenzy in the construction of data centers, the infrastructure that powers this technology. While China’s DeepSeek has demonstrated that AI models can be developed with fewer resources, U.S. President Donald Trump, in his first week in office, unveiled an ambitious plan to mobilize $500 billion over four years to build data centers and ensure they have a reliable energy supply.

However, the rapid growth of this market in the U.S. is beginning to strain the country’s energy grid, which is now seeking creative solutions — including the authorization of small modular nuclear reactors — to meet rising demand. But the dark side of AI production extends beyond its voracious energy and water consumption, which is used to cool the equipment, or the growing mountain of electronic waste it generates. It also poses risks to public health, a concern the White House recently acknowledged for the first time.

One of the latest executive orders signed by former President Joe Biden (EO 14141, January 14), titled Advancing United States Leadership in Artificial Intelligence Infrastructure, implicitly acknowledges that data centers are harmful to health. Unlike Biden’s executive order aimed at reducing the risks posed by AI to citizens and national security — which was overturned by Trump — this order has not been revoked. It ensures the transfer of federal land for the construction of data centers, which are deemed “strategic” for the country’s economy and security. The order also outlines the development of sufficient energy resources, including nuclear energy if necessary, to power these facilities.

One section of the executive order details the requirements for the locations where these new data centers will be built. These include factors such as topographically suitable soil, respect for surrounding wildlife and cultural resources, proximity to high-voltage networks, and transportation connections. It also highlights another fact: “Location within geographic areas that are not at risk of persistently failing to attain National Ambient Air Quality Standards, and where the total cancer risk from air pollution is at or below the national average according to the Environmental Protection Agency’s (EPA’s) 2020 AirToxScreen.”

“The executive order acknowledges for the first time the public health impact of data centers,” says Shaolei Ren, associate professor of electrical and computer engineering at the University of California, Riverside and an expert on AI sustainability. “While the order is likely to be amended or revoked, the explicit recognition of declining air quality and rising cancer rates highlights the immediate risk of AI data centers.”

Ren and colleagues have written a scientific paper, currently under review, on the economic and human costs of diseases directly linked to pollution associated with data centers. The primary issues stem from toxic gases, such as nitric oxide and PM2.5 particles, released during the electricity generation process that powers these facilities. These pollutants come from both the power plants, typically located just a few miles from the data centers, and from the backup generators used when the primary power supply is interrupted.

The figures Ren presents are substantial: the health costs attributed to AI-related pollution ranged from $17 billion to $29 billion between 2019 and 2023. During this period, there were at least 1,100 premature deaths, according to his calculations.

Data centers are neither more nor less polluting than other industries. The difference, according to Ren, is the rapid growth of the sector and the apparent lack of awareness about the pollution it generates. “It is well known that cars pollute, and, precisely for that reason, there are strict regulations to control and reduce the gases they emit. But data centers are growing at such a dizzying rate that by 2028 they will exceed the emissions of the entire California vehicle fleet — even if we add 35 million vehicles — according to the Department of Energy’s recent projection of data center energy demand.”

What has the Trump administration said about this? Nothing directly, but there have been some signals. The first, and most significant, is that it has not revoked Biden’s executive order, meaning it has not explicitly rejected its content. The second signal came this week: the new head of the EPA, Lee Zeldin, introduced an initiative on Tuesday titled Powering the Great American Comeback. While the initiative doesn’t mention carbon dioxide or other greenhouse gases, it positions “clean air” as the first pillar of the effort to “protect human health and the environment.” It also emphasizes the importance of ensuring that AI data centers “can be powered and operated in a clean manner with American-made energy.” According to Ren, “In this context, the term ‘clean’ seems to refer to low emissions of air pollutants.”

Where to build what type of facility

This report offers another clue: more than half of the data centers in the U.S. by 2028 will be colocation centers. In industry terminology, these are facilities that host third-party data and, unlike those developed for large technology companies — known as hyperscale data centers — ”are typically located in urban areas with high population density,” Ren explains. “These data centers, along with those specializing in AI, which have much higher energy demands than the rest, and facilities located near downwind populations, will have the greatest impact on public health,” says the professor.

Biden’s executive order aims to minimize the health impact of data centers by strategically determining where certain tasks should be performed. For instance, the most energy-intensive and, therefore, most polluting processes, such as model training, could be carried out in facilities located in desert areas, while less energy-demanding tasks could be assigned to data centers in densely populated areas. “The proposal is to prioritize data center projects in locations where air quality is good and cancer rates from air pollutants do not exceed the national average. In other words, the goal is to avoid further exacerbating public health inequalities,” Ren explains.

A recent study conducted by the scientist’s team used the locations of Meta’s data centers in the U.S. as a case study to analyze the potential health impacts of following this decision-making approach — specifically, training models in areas with low health impact, such as sparsely populated regions. The results indicate that, based on Meta’s reported activity in 2023, the health costs — i.e., medical expenses related to respiratory diseases and cancer caused by pollution — could be reduced by 30%.

Sign up for our weekly newsletter to get more English-language news coverage from EL PAÍS USA Edition