Reviewed by Lexie CornerNov 4 2024
Researchers from Zhejiang University and Nankai University conducted a comprehensive assessment of carbon emissions from artificial intelligence (AI) systems, published in Frontiers of Environmental Science & Engineering.
Top 20 AI systems in terms of carbon emission for a single training run. Image Credit: Frontiers of Environmental Science & Engineering
As AI technology advances, the energy demands for training complex models have grown significantly, raising concerns about associated carbon emissions.
Driven by growing demand from both academia and industry, this rapid expansion is fueling exponential increases in processing power but with notable environmental consequences. Addressing these challenges requires further research to fully understand AI’s carbon footprint and develop strategies to mitigate its environmental impact.
The study, which analyzed emissions from 79 major AI systems from 2020 to 2024, underscores the need for standardized emissions limits and regulatory measures. The findings indicate that these AI systems could collectively emit over 102 million tons of CO₂ per year, highlighting an urgent need for regulations to reduce AI’s environmental impact.
The research team analyzed carbon emissions from 79 leading AI models published between 2020 and 2024 and uncovered significant differences in energy consumption. For example, Google’s Gemini Ultra model alone accounts for 36.7 % of emissions among top-performing AI systems, while GPT-4’s emissions rose twelvefold compared to its predecessor.
With the global demand for AI services rising, the study reveals that operational demands often exceed emissions from initial training, with annual emissions estimated to be 960 times those of a single training run.
The economic impact is substantial: AI-related emissions could cost the industry over $10 billion annually, assuming a carbon price of $109 per ton. As AI’s carbon footprint now matches the annual emissions of entire countries, these findings underscore the need for standardized emissions metrics and limits.
The exponential growth in AI capabilities mirrors a concerning rise in its environmental impact. This study underscores the urgent need for the AI industry to adopt greener practices and sustainable standards. Our goal is to equip policymakers with the data needed to address AI’s carbon footprint through proactive regulations.
Dr. Meng Zhang, Lead Researcher, Zhejiang University
Dr. Zhang emphasized that achieving a sustainable future hinges on balancing environmental responsibility with AI innovation.
The study’s findings carry significant implications for both AI development and environmental policy. As AI applications expand globally, controlling carbon emissions becomes essential for minimizing environmental impact and achieving climate goals.
Emission caps could drive the industry toward energy-efficient practices, fostering innovation in sustainable AI technology. Additionally, AI-related carbon metrics could help policymakers establish effective standards, ensuring that the environmental costs of AI development remain manageable.
This study was financially supported by the National Key Research and Development Program of China (No. 2022YFC3203003), the Fundamental Research Funds for the Central Universities (No. 226-2024-00010), and the Key Project of Natural Science Foundation of Zhejiang Province (No. LZ23E080004).
Journal Reference:
Yu, Y. et. al. (2024) Revisit the environmental impact of artificial intelligence: the overlooked carbon emission source? Frontiers of Environmental Science & Engineering. doi.org/10.1007/s11783-024-1918-y