Actual emissions could exceed the Harvard study’s estimate

Guidi, a Harvard University postdoctoral research fellow focusing on AI and public health, said preliminary results from newer studies identify around 1,000 hyperscale data centers in the United States, more than double the 403 facilities covered by his study. In the PBS News report published November 17, 2026, he confirmed that actual emissions could be much higher than the study’s estimate. PBS also reported that Virginia, Iowa, Oregon and Ohio account for half of the electricity used by data centers.

The Harvard study covered May 2024 through April 2025 and found that data centers accounted for nearly 2 percent of total U.S. energy consumption. It estimated that U.S. data centers produced more than 25 million metric tons of carbon dioxide, a level two or three times higher than an earlier estimate in 2018.

The PBS account said the 400 largest U.S. data centers emitted more greenhouse gas emissions into the air in one year than the entire nation of Denmark. Guidi elaborated that the study’s 403 facilities were just 400 buildings, yet their annual carbon emissions were comparable to those of an entire country with millions of people, homes, hospitals and industries. He described the scale as extraordinary and said the facilities represented only a fraction of the industry.

Beyond the facility count, Guidi said the actual scope of emissions could be much higher than his report suggests. He coupled that possibility with water consumption and potentially public-health consequences, saying the technology is moving faster than the ability to understand its consequences.

The Harvard study did not directly address public health. Guidi said power generation at the plant level produces carbon emissions as well as hazardous pollutants that travel through the air and can be inhaled, and he described the resulting health effects as dangerous. Guidi also said there is plenty of scientific research on the health effects of these pollutants.

Virginia, Iowa, Oregon and Ohio account for half of the electricity used by data centers, according to the PBS report. Guidi said those states provide the infrastructure, land and electricity that large facilities need. He said the benefits of the expansion are not necessarily felt by the same communities that bear its environmental costs, and more electricity demand can mean additional power plants, pollution and pressure on local infrastructure. “So those consequences don’t stop at the data center’s fence,” he said.

Asked about arguments that AI data centers are needed for economic competitiveness and national security, Guidi said, “Yes, I see that, and there’s a big backlash happening right now.” He said he would not call the facilities a necessary evil. As an AI researcher, Guidi said he is excited about the technology and what it can do, but said better ways to build it might cost more and take longer. “The question is whether we’re willing to prioritize doing it right over doing it first,” he said.

Guidi identified water consumption, electricity demand, strain on the power grid and potential public-health consequences as the main concerns. The PBS account also said researchers could not capture all available information from many data centers.

By the time of the interview, the report was several months old and the study’s data were already more than a year old, Guidi said. He said scientific evidence should not arrive after the infrastructure it is meant to examine, and neither should public participation. “We’re building this AI infrastructure faster than we can independently understand its consequences,” he said.