X user @kane criticized the New York Times over a country comparison for data-center emissions. In a November 19 quote-post of @dwallacewells, who had linked to the newspaper’s coverage, the account called the NYT and its readers “largely innumerate” and offered national percentages instead. Source Source
In the November 17 post @kane quoted, @dwallacewells shared a NYT link alongside a quotation describing an estimated 25.7 million metric tons of annual carbon dioxide emissions from the largest U.S. data centers as “almost as much as entire countries like Bolivia or Denmark.” Source The full NYT article wasn’t available for inspection.
@kane countered: “25.7 MT CO2 is 0.4% of US emissions. 0.16% of China emissions.” Source
Percentages don’t get a context exemption. The post’s text doesn’t identify the national totals, comparison years or emissions-accounting definitions behind those figures. Those details matter when comparing carbon dioxide with totals that might include other greenhouse gases. The percentages haven’t been independently verified for this article. Source
A footprint can resemble one country’s annual emissions while representing a small fraction of a much larger country’s total. Those comparisons answer different questions; neither, by itself, establishes how environmentally consequential the footprint is.
What Harvard’s estimate actually covers
According to Harvard T.H. Chan School of Public Health’s November 16 release, the Harvard-led study was published that day in Nature Sustainability. Researchers used satellite imagery to identify 403 U.S. hyperscale data centers—the largest class of facilities supporting services including cloud computing and AI—and estimated their electricity consumption from May 2024 through April 2025. Source
They then used Environmental Protection Agency data on regional power-generation mixes to estimate the associated emissions: about 25.7 million metric tons of CO2. That’s a modeled footprint tied to electricity use, not a direct measurement of emissions coming out of 403 buildings. Source
Depending on operating assumptions, the facilities used an estimated 68–99 terawatt-hours of electricity. Harvard says their demand represented approximately 1.8% of total U.S. electricity consumption during the study period. That is an electricity share, not the emissions share @kane cited. Roughly 54% of the electricity was attributed to fossil-fuel generation. Source
Researchers lacked enough data to include every U.S. hyperscale facility, and they couldn’t separate general-purpose computing from AI-specialized operations. Calling the entire footprint AI’s carbon bill would hand it the whole tab without establishing its share. Source
Nor is this a November 2026 emissions measurement. The researchers believe electricity demand and emissions are likely higher now, but Harvard’s release doesn’t quantify a new total. Lead author Gianluca Guidi said in the release: “What we’re seeing is already striking, but it’s only a snapshot of a rapidly expanding industry.” Source
Comments