Ilaria Rosella Pagliaro Study estimates that by 2030, up to 5 million tonnes of e-waste could be generated by AI, highlighting the need for sustainable solutions
Artificial Intelligence is rapidly perfecting its capability to mimic human creators. Today, generative AI can converse with people, create art, make films, and even teach itself how to replicate video games. But a new study by researchers from the Chinese Academy of Sciences and Reichman University in Israel suggests a growing concern: AI might also reproduce one of modernity’s less noble traits-the tendency to harm the environment.
The explosion of generative AI systems, of which chatbots like ChatGPT and other content-creation tools are a part, could add anywhere between 1.3 and 5.5 million US tons of electronic waste (e-waste) by 2030. The study zeroes in on large language models, or LLMs-tools engineered to process and produce humanlike text by interpreting the complex statistical relationships that underlie the use of language. While these systems offer pragmatic and creative advantages, they also bring more abstract and real-world dangers with them, like job losses and manipulative risks. Now, it seems they contribute to a major environmental concern, too.
Large language models and the infrastructure burden
Large language models need computing infrastructure that is both powerful and complexly fitted with advanced chips. The constant upgrading that this growing technology would demand now threatens to add to the already-out-of-hand e-waste problem.
“LLMs require large computational resources for training and inference, which in turn comes at high energy costs and significant carbon footprint,” the study points out.
Previous research has usually focused on energy consumption and carbon emissions generated, but overlooked the physical materials required in the lifecycle of these systems and electronic waste resulting.
According to an estimate by Peng Wang, a researcher at the Chinese Academy of Sciences, AI-related e-waste could top 5.5 million US tons by 2030—equal to every person discarding a smartphone. In that extreme scenario, the waste would include 1.7 million US tons of electronic boards and 550,000 US tons of batteries, which are potentially harmful to the environment.
The research describes four scenarios of AI’s future, with different levels of AI take-up; the most advanced scenario sees a rise in e-waste to 2.8 million US tons annually by 2030. That would contribute to the general increase of technology-related waste, which is projected to grow 30% to reach 90 million US tons worldwide.
Solutions to reduce AI’s environmental impact
Despite these alarming figures, the study identifies effective ways to mitigate this environmental burden. Circular economy strategies, supported by the International Energy Agency and numerous companies, propose solutions like extending the lifespan of components and reusing materials in production phases. These measures could reduce AI’s contribution to e-waste by as much as 86%.
Source: UNITAR
