E-waste from AI computers could 'escalate beyond control'
October 28, 2024
The increasing popularity of generative AI is projected to result in the rapid growth of e-waste, electronic waste, according to a study published in Nature Computational Science .
Researchers behind the study calculated that e-waste could reach a total of 1.2-5.0 million metric tons between by 2030, which is around 1,000 times more e-waste than was produced in 2023.
"We found that the e-waste generated by generative AI, particularly large language models, could increase dramatically — potentially reaching up to 2.5 million tons per year by 2030 if no waste reduction measures are implemented," said Asaf Tzachor, an expert in sustainable development at Reichman University in Israel, and a co-author of the study.
The study also offers solutions to reducing e-waste — strategies to prolong, reuse, and recycle generative AI hardware could reduce e-waste creation by 16% to 86%, they estimate.
"This presents a tremendous opportunity for reducing the waste stream if these practices are widely adopted. It's clear from this study that the nature of the e-waste crisis is global, which is why it's important to focus on cross-border e-waste management," said Saurabh Gupta, founder of Earth5R, an India-based sustainability organization. Gupta was not involved in the study.
What is e-waste?
Every time we throw away an 'outdated' or broken electronic device, it's considered e-waste. This can include computers, smartphones, chargers and wires, electronic toys, cars, and larger server systems.
E-waste makes up 70% of the total toxic waste produced around the world each year, yet only 12.5% of e-waste is recycled. This live counter at The World Counts shows just how fast e-waste is growing.
"Reducing e-waste is important because improper disposal leads to the release of hazardous materials, like lead and mercury, which harm ecosystems and human health," Gupta told DW via email.
The researchers in the study published October 28, 2024, focused on e-waste produced from generative AI algorithms — types of AI that generate texts, images, videos or music from massive datasets.
It's clear from previous research that AI has high energy requirements — calculations by research firm SemiAnalysis suggest AI could result in data centers using 4.5% of global energy production by 2030.
But Tzachor said it's less clear how much e-waste is produced from generative AI programs, such as ChatGPT. This includes all the computer resources required for training and using AI in data centers.
And because generative AI is dependent on rapid improvements in hardware infrastructure and chip technologies, there are indications it's leading to more e-waste as the hardware gets updated or replaced.
"It's far easier and more cost-effective to address the e-waste challenges posed by AI now, before they escalate beyond control," said Tzachor.
How did researchers calculate the growth in AI e-waste?
The researchers created a model to quantify the scale of e-waste from data centers that support the use of generative AI models, such as large language models.
They found that e-waste could reach 5 million tons per year in a scenario where growth AI was estimated to be high.
But their estimates of AI e-waste were potentially on the low side, said Tzachor, because of the rapidly changing AI business landscape.
"Factors such as geopolitical restrictions on semiconductor imports and rapid server turnover may intensify the generation of e-waste associated with generative AI," Tzachor told DW via email.
Moreover, the study only included e-waste created by generative AI systems, specifically large language models, and not other forms of AI.
"E-waste from the broader AI ecosystem is significant. The study forecasts that this figure will rise with increasing AI adoption, creating a combined environmental challenge from multiple forms of AI," said Gupta.
Reducing e-waste needs global strategies
The study estimates that implementing circular-economy strategies could reduce e-waste generation by 16%, or up to 86%.
Circular economy strategies aim to minimize waste and increase the efficiency of computer hardware.
Tzachor said there were three main goals of the strategy:
- Prolong the use of existing hardware to delay the need for new equipment
- Reuse and remanufacture components
- Extract valuable materials during recycling of hardware
Gupta said he strongly agreed with the study's findings.
"The range of 16-86% reduction reflects the immense potential of these strategies, especially if supported by policies, and when widely implemented across industries and regions," said Gupta.
Gupta's organization, Earth5R, has demonstrated how effective circular economy strategies approaches can be, he said.
"Through our grassroots programs and partnerships with businesses, we are already fostering local e-waste collection and recycling efforts that help businesses and consumers manage their electronics sustainably," said Gupta.
He emphasized that e-waste was a global crisis that needed equitable, cross-border e-waste management strategies to mitigate the "environmental and health damage" caused when high-income countries export their e-waste to low-income regions.
Edited by: Zulfikar Abbany
Primary source:
E-waste challenges of generative artificial Intelligence, published by Wang, P et al. in the journal Nature Computational science (October 2024) DOI: 10.1038/s43588-024-00712-6