As debates about the environmental cost of artificial intelligence intensify, a new peer-reviewed study offers evidence that sharply contrasts with public fears. While the global rise of data centers and large-scale AI systems has fueled concerns about soaring electricity use, a landmark analysis published in Environmental Research Letters suggests that AI’s real environmental footprint—based on current adoption levels—remains surprisingly small. The study, “Watts and Bots: The Energy Implications of AI Adoption,” authored by Anthony R. Harding and Juan Moreno-Cruz and published in November 2025, presents one of the most comprehensive evaluations of how AI affects national energy consumption and carbon emissions.
Harding and Moreno-Cruz approached the issue through a wide economic perspective. Rather than examining only the direct electricity used by AI systems, they analyzed how AI influences productivity and output across multiple sectors, modeling the indirect energy demands that follow. By integrating projections of AI use with national economic and energy-intensity data, the researchers calculated the additional energy required if AI adoption continues at its observed pace.
AI Environmental Impact Not Too Much
The results reveal that AI’s overall contribution to national energy use is far smaller than popularly assumed. The study estimates that current AI adoption would add roughly 28 petajoules of energy consumption per year and produce about 896 kilotons of CO₂ emissions. In national terms, this equals just 0.03 percent of annual U.S. energy use and 0.02 percent of carbon emissions—numbers that position AI as a comparatively minor factor in the national climate landscape.
Lead author Anthony R. Harding said the findings demonstrate how public narratives have often overstated the danger. “There is a clear mismatch between perception and reality,” Harding stated.
“People imagine AI as an enormous, constantly expanding energy sink, but the data simply doesn’t support that. At current adoption levels, AI is nowhere near becoming a major driver of national emissions.” He added that policy discussions “must rest on empirical evidence, not on anxiety over the rapid visibility of new technologies.”
Co-author Juan Moreno-Cruz agreed that the results counter common fears but cautioned that the picture could change in the future. “Our conclusion is not that AI will never pose an environmental challenge,” he said.
“It’s that, right now, its contribution is very small. We are still in a ‘safe zone,’ but that doesn’t mean we should stop paying attention.” Moreno-Cruz emphasized that energy impacts could grow if AI systems become significantly larger or if adoption accelerates beyond current projections. Still, he stressed that many speculative claims circulating in public debate are not grounded in evidence.
From mountains to oceans, delivered to you. Follow us on Lingkar Bumi WhatsApp Channel.
The study finds that the environmental impact of AI varies by industry. Sectors with deep reliance on computational infrastructure—such as information services or advanced manufacturing—could experience more substantial energy increases, with some scenarios projecting up to 12 petajoules of added consumption and as much as 272 kilotons of CO₂ annually. Even then, these figures remain minor compared with the emissions produced by transportation, heavy industry, and fossil-fuel combustion. Harding noted that “AI’s energy footprint is dwarfed by traditional high-emission sectors, yet AI often receives disproportionate scrutiny because it is new and highly visible.”
Harding and Moreno-Cruz also highlighted that localized environmental effects can be more pronounced than national ones. Communities hosting large data centers may face higher demands on local grids, infrastructure pressures, or spikes in electricity prices.
“Local strain is real,” Harding said, “but it should not be confused with national-scale climate impact.” The authors argue that such local issues are more a matter of energy planning and grid management than global environmental risk.
Important Tools to Mitigating Climate Impact
The researchers also point out that AI could become an important tool in mitigating climate impacts rather than worsening them. In their analysis, they emphasize the potential of AI to optimize energy use across sectors such as transportation, logistics, grid management, and industrial operations.
“If deployed responsibly,” Moreno-Cruz said, “AI can help us reduce waste, forecast energy demand more accurately, and improve efficiency in ways that offset its own consumption. AI can be part of the solution—not just part of the problem.”
The paper acknowledges certain limitations, including reliance on energy-intensity data from 2014. Changes in technological efficiency, grid decarbonization, or shifts in AI hardware could influence real-world energy trends. Still, the authors argue that the overall conclusion remains clear, under current adoption patterns, AI is not a major contributor to national energy use or emissions.
As public concern over AI’s environmental impact continues to grow, the study provides a grounded and data-driven counterweight to speculation. Harding summarized the message succinctly. “Right now, AI’s climate impact is modest. The conversation should reflect that reality—while still keeping an eye on what may come next.” (Sulung Prasetyo)
