A new Cornell University study published in Nature Sustainability provides the first state by state accounting of the environmental footprint from AI data centers. The research projects that by 2030, U.S. data centers powering AI could emit 24 million to 44 million metric tons of CO2 annually, equivalent to adding 5 to 10 million cars. Water consumption could reach over 1 billion cubic meters per year, enough for 10 million households. The study identifies siting decisions as the single most impactful lever: locating facilities in the Midwest and windbelt states, combined with grid decarbonization and liquid cooling, could cut carbon emissions by roughly 73 percent and water use by 86 percent. Even with ambitious renewable energy buildout, the researchers warn that total emissions could still rise 20 percent if AI demand grows faster than the clean energy transition. Reaching net zero would require 28 gigawatts of new wind or 43 gigawatts of solar capacity. The article also covers New Jersey specific developments like CoreWeave's 140 megawatt facility in Kenilworth and a proposed 300 to 400 megawatt hyperscale site in Vineland. Senator Bernie Sanders has introduced legislation for a public ownership stake in major AI companies to address environmental and social concerns. The study underscores that infrastructure choices made this decade will determine whether AI becomes a climate accelerant or a burden.
