Researchers examined how the spread of artificial intelligence (AI) data centers in the United States affects the environment and the economy. They looked at three main drivers: growing electricity demand, the extra cooling AI equipment needs, and how backup power systems operate. Their analysis finds that the impacts are shaped less by a single facility’s design and more by the broader electricity, water, and land‑use systems where the facility sits.

The team compared traditional data centers with AI facilities and found clear differences. AI workloads use large clusters of specialized processors such as GPUs. These clusters draw more power per area and produce more heat than conventional computing. A typical data center can use as much electricity as about 25,000 households, and new requests to connect large facilities to the grid now often exceed 100 megawatts and can approach 1 gigawatt. Projections cited in the paper put U.S. data center electricity use at roughly 4.4% of total U.S. electricity in 2023 and potentially rising to 6.7%–12% by 2028 or 9%–17% by 2030, depending on hardware and cooling choices.

Most environmental impacts come from electricity use. The authors emphasize marginal emission factors — that is, the change in power‑sector emissions when demand rises — because incremental demand is usually supplied by the generators called on at the margin. In many U.S. regions, those marginal generators are fossil‑fuel plants, mainly natural gas and sometimes coal, so extra electricity often adds greenhouse gases and air pollutants. The paper notes coal produces the highest emission intensities, natural gas lower values for some pollutants, and diesel backup generators produce especially high levels of nitrogen oxides, carbon monoxide, and particulate matter. The authors also report that cooling systems can consume significant water in some sites, and that noise and land‑use changes are local effects that vary by region.