AI adoption in US adds ~900k tons of CO₂ annually, study finds (techxplore.com)

A new study in Environmental Research Letters estimates that broad AI adoption across the U.S. economy would add roughly 896,000 tons of CO₂ per year—about 900,000 tons—equivalent to roughly 0.02% of current U.S. emissions. The authors modelled AI integration across sectors to project increased energy demand and carbon output, finding industry-level energy use could rise by as much as 12 petajoules annually (roughly the electricity consumption of 300,000 U.S. homes). The report frames the increase as measurable but modest relative to other industrial sources, while noting the aggregate impact is nontrivial as AI scales. For the AI/ML community, the paper highlights a practical takeaway: emissions from AI are not negligible and will grow with wider deployment, so engineering and product decisions matter. Technical implications include prioritizing energy-efficient model architectures, server and data-center optimization, workload scheduling, and sector-specific deployment strategies to minimize marginal energy use. The authors urge integrating sustainability metrics into development and procurement processes to ensure responsible scaling. (Watts and Bots: The Energy Implications of AI Adoption, Environmental Research Letters, 2025; DOI: 10.1088/1748-9326/ae0e3b.)