A Study on the Impact of Generative Artificial Intelligence Growth on Ecosystem Material Cycles: Analyzing Resource Use, Exergy Use, and Greenhouse Gas Emissions

The rapid advancement of artificial intelligence (AI) technology is driving transformative changes across society. However, this progress also entails significant resource and energy demand, posing substantial new challenges to the Earth’s ecosystems. Specifically, the environmental impacts arising from AI model training and inference, data center operations, and the manufacturing and disposal of electronic devices threaten the balance of ecosystem material cycles and could exacerbate climate change. Therefore, it is urgently needed to understand the effects of generative AI technology growth on ecosystem material cycles and to identify sustainable AI technology development and application strategies. This study aims to quantitatively assess the resource consumption (including metals, plastics, and water), exergy use (primarily through electricity demand and fossil fuels), and greenhouse gas emissions associated with the anticipated growth of generative AI technology and its consequent impacts on ecosystem material cycles. First, we analyze resource and exergy use within the generative AI industry, encompassing AI model training and inference, data center operations, and the production of AI chips and devices. We quantify the consumption of key elements and water, alongside the exergy demand for electricity and fossil fuels. We employ a Life Cycle Assessment (LCA) methodology to evaluate the comprehensive environmental footprint of AI technology. Second, we examine the environmental impact of AI-related waste by evaluating the generation, treatment processes, and ecosystem effects of electronic waste (including AI chips, devices, and data center equipment). This analysis focuses on the environmental leakage pathways of hazardous and plastic waste and the patterns of material movement within the ecosystem, particularly with regards to soil and water pollution and biodiversity loss. Third, we model the impact of generative AI technology on key ecosystem material cycles, such as carbon, nitrogen, and phosphorus. We estimate changes in resource use, exergy consumption, and waste generation under multiple AI technology growth scenarios. Finally, we propose strategies for the sustainable development and application of AI technologies. Based on our findings, we will formulate concrete policy and technical recommendations for developing and implementing resource-efficient and low-exergy-consuming AI technologies.