Ellen Phiddian
Cosmos science journalist
The explosion in generative AI could be responsible for an extra 5 million tonnes of e-waste by 2030, according to a new study.
The report, published in Nature Computational Science, estimates that without changes to regulation, generative AI technology alone will produce 1.2-5.0 million tonnes of e-waste by the end of the decade.
The study also finds that annual generative AI e-waste could increase a thousandfold, rising from 2,600 tonnes in 2023 to as much as 2,500,000 tonnes per year in 2030.
Circular economy practices could reduce this waste by 16%-86%, according to the analysis.
“Previous studies on sustainable computing have primarily focused on the energy use and carbon emissions of AI models,” write the researchers in their paper.
“However, the physical materials involved in their life cycle, and the waste stream of obsolete electronic equipment – known as electronic waste (e-waste) – have received less attention.”
Rapid advances in computer chips that facilitate AI are making older hardware obsolete very quickly, according to the researchers.
The team used computer models to estimate the amount of hardware used and discarded by generative AI from 2020-2030.
They modelled both a conservative scenario, where the technology was only used for specific applications, and an aggressive scenario, where it was used widely.
In the aggressive scenario, annual e-waste production rose to 2.5 million tonnes per year in 2030, and 5.0 million tonnes total.
This includes 1.5 million tonnes of printed circuit boards and 0.5 million tonnes of batteries, both of which contain hazardous, polluting materials.
In the conservative scenario, annual production rose to 0.4 million tonnes per year by 2030, and 1.2 million tonnes total.
“For context, the most recent Global E-waste Monitor report indicates that annual e-waste related to small information technology equipment such as personal computers totalled 4.6 million tonnes in 2022, and will sum up to 43.2 million tonnes by 2030, meaning that AI servers could increase this quantity by 3%-12%,” write the researchers.
They also point out that this amount of e-waste equates to every person on the earth in 2030 throwing out between 0.2-1.6 iPhones.
The researchers also analysed the potential of 3 circular economy strategies – improving performance, increasing lifespan, and re-using modules in manufacturing – to reduce this waste.
While the effects of these strategies may be complicated – for instance, making chips higher-performing could trigger more demand – the researchers calculate that e-waste could be reduced by up to 86% if these options are implemented across the whole supply chain.
The researchers also point out that geopolitical tensions shaping the semiconductor industry could exacerbate e-waste. If countries can’t buy the latest chips for their AI technology, they’ll use older models that will be disposed of faster.
“Our findings underscore the need to acknowledge the potential for future swells of generative AI-related e-waste and to proactively implement circular, design and management strategies to avoid them,” write the researchers.
