GENERATING A(I) CRISIS

The document argues that the rapid growth of generative AI is driving a significant increase in electricity demand, which threatens climate goals by incentivising the continued use of fossil fuels and diverting clean energy resources. It highlights a lack of transparency in corporate emissions reporting, where 'market-based' accounting masks actual grid impacts, and warns that 'moonshot' technology promises like nuclear fusion may act as mitigation deterrence. Specific examples include data centre growth in Ireland and Malaysia, and the rising emissions profiles of companies like Microsoft and Google.

• Projections for power demand from machine learning vary widely due to different assumptions about training and usage. The IEA projects dedicated machine learning power demand will be 10x greater in 2026 than in 2023, while a Goldman Sachs report suggests data centres could grow from 1-2% to 3-4% of global power demand, and up to 8% of US power demand.
• The growth of data centres is actively competing with other electrification needs and may lead to increased fossil fuel reliance. In Ireland, all new clean energy in 2023 served new data centre growth, leaving none for other industries, and 20 recent requests from data centres were for direct connections to the fossil gas grid.
• Large tech companies are using 'market-based' accounting to report lower emissions than their actual grid usage suggests. This method relies on renewable energy certificates (RECs) or power purchase agreements (PPAs) that may not incentivise new renewable energy or match the time and location of actual power consumption.
• Major technology firms are seeing their climate targets undermined by AI growth. Microsoft's president admitted its carbon ambitions have shifted because "the moon has moved" due to machine learning investments, and Google's reports show a rise in power consumption and a shift toward carbon removal investments rather than absolute reductions.
• The document identifies a risk of 'mitigation deterrence,' where improbable 'moonshot' technologies like nuclear fusion or carbon capture are promoted as solutions, potentially deterring investment in feasible renewable energy. An example is a plan in Salt Lake City to replace coal with next-generation nuclear plants that was shelved in favour of prolonging coal burning.
• Generative AI is being used to amplify climate disinformation. In Australia, an anti-wind farm group used machine learning to create a fake scientific paper to spread via Facebook.
• Data centre growth is particularly concerning in the Asia-Pacific (APAC) region, where grids are more reliant on fossil fuels. In Malaysia, projected electricity use by data centres could exceed the country's total renewable output from 2022.

APA

Climate Action Against Disinformation (n.d.). GENERATING A(I) CRISIS. https://caad.info/wp-content/uploads/2024/12/GENERATING-AI-CRISIS-The-impact-of-generative-AI-and-machine-learning-systems-on-corporate-accountability-and-power-system-decarbonisation.pdf

Chicago

Climate Action Against Disinformation. GENERATING A(I) CRISIS. n.d. https://caad.info/wp-content/uploads/2024/12/GENERATING-AI-CRISIS-The-impact-of-generative-AI-and-machine-learning-systems-on-corporate-accountability-and-power-system-decarbonisation.pdf.

Wikipedia

{{cite report |author=Climate Action Against Disinformation |title=GENERATING A(I) CRISIS |url=https://caad.info/wp-content/uploads/2024/12/GENERATING-AI-CRISIS-The-impact-of-generative-AI-and-machine-learning-systems-on-corporate-accountability-and-power-system-decarbonisation.pdf |access-date=23 October 2026 |via=Climate Insights Directory}}

BibTeX

@techreport{climateactionagainstdisinformationndgenerating, author = {{Climate Action Against Disinformation}}, title = {{GENERATING A(I) CRISIS}}, institution = {Climate Action Against Disinformation}, url = {https://caad.info/wp-content/uploads/2024/12/GENERATING-AI-CRISIS-The-impact-of-generative-AI-and-machine-learning-systems-on-corporate-accountability-and-power-system-decarbonisation.pdf}, urldate = {2026-10-23}, note = {Indexed by Climate Insights Directory} }