Study finds 74% of AI climate claims lack evidence
A report released on February 17 by climate nonprofits Beyond Fossil Fuels and Climate Action Against Disinformation analyzed 154 industry and institutional assertions regarding AI’s climate advantages. The study, authored by energy analyst Ketan Joshi, determined that 74% of these claims lacked verifiable evidence. Researchers warn that the energy-intensive nature of generative AI may actually increase global emissions.
The analysis examined statements from technology companies and institutional sources. It found that the environmental benefits frequently cited by proponents are largely linked to older, leaner forms of machine learning rather than the generative AI tools currently driving data center expansion. The report identified no instance where consumer-facing applications such as Google Gemini or Microsoft Copilot resulted in a material, verifiable reduction in emissions.
Researchers scrutinized an International Energy Agency report that was reviewed by industry giants. According to the study, the IEA’s assessment of AI’s climate impact showed an even split between academic sources, corporate materials, and unsubstantiated claims. The projection frequently cited by the industry—that AI could reduce 5 to 10% of global greenhouse gas emissions by 2030—was found to be “contested.”
Ketan Joshi characterized the industry’s focus on climate solutions as a diversion from the environmental impact of data centers. “What we see is companies veering wildly away from their climate targets,” Joshi wrote. “This focus on ‘AI for climate’ is a distraction from the decision to worsen the pollution of data centres through an unprecedented explosion of digital bloat.”
Major technology companies are facing increasing scrutiny over the environmental consequences of their AI investments. Google reported a 48% increase in greenhouse gas emissions between 2019 and 2023, attributing the rise largely to data center energy consumption. Microsoft acknowledged a 29% increase in emissions since 2020, citing the construction of facilities “designed and optimized to support AI workloads.”
Operational data indicates a significant rise in energy demand associated with AI tools. Research shows that a single ChatGPT query consumes approximately ten times the electricity of a traditional Google search. Goldman Sachs estimates that data centers will account for 8% of total U.S. power usage by 2030, increasing from 3% in 2022, a shift driven primarily by AI proliferation.
The coalition behind the report, which includes Stand.earth, Friends of the Earth U.S., and the Green Web Foundation, is calling for increased transparency regarding energy use and accountability for unverified environmental claims. They advocate for stricter reporting standards to address the disparity between reported emissions and the industry’s stated climate goals.