Published on
A new report is casting serious doubt over claims from some artificial intelligence (AI) companies that their products can meaningfully reduce carbon emissions.
ADVERTISEMENT
ADVERTISEMENT
Estimates of AI’s climate impact vary widely. A January study published in the journal Patterns found that data centres alone may have emitted between 32.6 million and 79.7 million tonnes of carbon dioxide in 2025, which is roughly equivalent to the annual emissions of a small European country.
Meanwhile, the International Energy Agency (IEA) has suggested that AI could reduce global emissions by up to 5 per cent by 2035 by speeding up energy sector innovations, potentially offsetting the emissions generated by data centres. For example, the IEA said AI could help scientists test materials and battery chemistries to support new solar power technology.
Beyond Fossil Fuels, a German non-profit, examined over 150 climate-related claims from the world’s biggest AI companies and organisations like the IEA to see what type of evidence supports the claims that AI could cut emissions.
Only 26 percent of their sample cited published academic papers to support their claim, and another 36 percent did not cite any evidence. The remainder leaned on corporate reports, media articles, NGO publications or unpublished academic work.
The analysis notes that corporate sources rarely include peer-reviewed evidence or primary data to substantiate their claims.
“The evidence for massive climate benefits of AI is weak, whilst the evidence of substantial harm is strong,” the report notes.
For example, Google claimed AI could cut 5 to 10 per cent of global greenhouse gas emissions by 2030 if the technology scales. Researchers traced this claim to a 2021 blog post from consulting firm Boston Consulting Group (BCG), which extrapolated that figure from their experience with clients.
The analysis describes the Google claim as “extrapolation of massive global climate benefits … on seemingly anecdotal evidence”.
Many AI companies argue that smaller, narrowly trained models, such as those trained on a single high-quality database, are better for the environment.
Yet, the researchers caution that the claims about narrow AI models could be overstated because there is a lack of peer-reviewed evidence that shows that these models can meaningfully reduce emissions.
The analysis also did not find a single example where generative AI systems such as ChatGPT, Gemini, or Copilot led to a “material, verifiable and substantial level of emissions reductions.”
“Even if these benefits are real, they are unrelated to – and dwarfed by – the massive expansion of energy use from the generative AI industry,” the press release added.
The authors note that the results do not mean AI technologies do not have any climate benefits, but say there is little evidence that AI reduces emissions enough to offset the energy it will take to run these systems.
Euronews Next contacted OpenAI, Microsoft, Google and the IEA for statements on how they cite climate-related estimates.
Read the full article here