AI may produce millions of tonnes of electronic waste by 2030, study finds

Rapid rise of generative artificial intelligence requires frequent upgrades to hardware and chip technology, leaving existing equipment obsolete

Generative artificial intelligence could create close to 1,000 times more electronic waste by the start of the next decade, a global analysis has found.

Without any waste reduction measures, e-waste from AI would see significant growth - from 2,600 tonnes in 2023 to 2.5 million tonnes in 2030, the scientists said.

That would be equivalent to the world's projected population in 2030 of 8.5 billion people throwing away nearly two iPhones each.

Do you have questions about the biggest topics and trends from around the world? Get the answers with SCMP Knowledge, our new platform of curated content with explainers, FAQs, analyses and infographics brought to you by our award-winning team.

The rapid rise of generative AI requires frequent upgrades to hardware and chip technology, often rendering existing electronic equipment obsolete. That equipment contains toxic metals including lead and chromium which are hazardous to health and the environment, as well as precious metals such as gold, silver and platinum that could potentially be recycled.

"Our findings indicate that this e-waste stream could increase, potentially reaching a total accumulation of 1.2 to 5 million tonnes during 2020 to 2030," the researchers from the Chinese Academy of Sciences and Reichman University in Israel said.

"This may be intensified in the context of geopolitical restrictions on semiconductor imports and the rapid server turnover for operational cost savings," they wrote in an article published in the peer-reviewed journal Nature Computational Science on October 28.

In March, the CEO of Nvidia, the American semiconductor designer dominating the AI chips market, said the weight of the company's graphics processing units (GPUs) had grown more than 40 times in two years.

"Two years ago ... it was 70 pounds [32kg], 35,000 parts. Our GPUs now are 600,000 parts and 3,000 pounds [1,360kg]. That is kind of like the weight of a carbon fibre Ferrari," chief executive Jensen Huang said of the specialised processors fundamental to AI that can perform complex calculations quickly and efficiently.

In the study, the researchers said the heavy weight of the latest Nvidia platform designed for large language model inference, training and data processing tasks showed that generative AI was a "substantial material-intensive sector".

"This rapid growth in hardware installations, driven by swift advancements in chip technology, may result in a substantial increase in e-waste and the consequent environmental and health impacts during its final treatment," they said.

Globally, they found that more than half of the untreated waste streams would be in North America - the United States and Canada - at 58 per cent, given the major cluster of AI data centres there. A quarter would come from East Asia, which includes China, South Korea and Japan, and 14 per cent from the European Union and Britain.

US restrictions on the sale of advanced GPUs to China and technical barriers could also hurt the environment because data centres in China "are forced to use outdated server models", they said.

"[This] can result in the loss of computational efficiency in countries that do not have access to such chips, resulting in higher physical server demand."

They compared Nvidia's advanced H100 chip, which was prohibited by US regulators from being sold to Chinese customers due to national security concerns, with its modified export version, the H800, which has reduced capabilities to comply with export regulations to China.

"For instance, the Nvidia H800's bandwidth efficiency is half that of the H100, necessitating double the number to achieve equivalent performance," the researchers said.

"Our analysis indicates that a one-year delay in obtaining the latest chips could result in a 14 per cent increase in end-of-service e-waste, cumulatively totalling 5.7 million tonnes from 2023 to 2030, higher than the global quantity of small information and communications tech waste in 2022."

In 2022, a record 62 million tonnes of e-waste was generated globally. Only 22.3 per cent of that was documented as having been properly collected and recycled, according to a global e-waste report by the United Nations in March.

It projected that under a "business as usual scenario", the collection and recycling rate will drop to 20 per cent in 2030 because of the huge growth of global e-waste generation.

According to the UN, many people recycle e-waste informally in developing countries. Some break down a discarded piece of equipment into marketable components while others extract secondary raw materials by burning, leaching and melting.

But these methods lack proper treatment standards and emit harmful substances such as acids, dioxins and furan. "It is not efficient, leads to significant resource loss and high environmental pollution and poses health risks for workers and the local community," the UN said.

To reduce e-waste generation, the research team suggested a circular economy strategy that could reduce such waste by an estimated 86 per cent globally.

It includes extending the lifespan of AI-related hardware, reusing obsolete elements like GPUs, CPUs and batteries in lower-grade applications, developing more efficient computing algorithms and increasing the computing efficiency of chips.

Study lead author Wang Peng, a professor in material circularity and sustainability at the Chinese Academy of Sciences' Institute of Urban Environment in Xiamen, said while the strategy could apply to countries developing AI, they might have different focus areas depending on their progress in the technology.

"For instance, the US should focus on being more responsible in manufacturing hardware and developing algorithms because it is at the forefront of AI development. The global AI industry could be more sustainable if the source is managed well," he said.

"As for China, the focus would be on strengthening regulation of operation and disposal processes. AI hardware is being manufactured in large quantities, followed by frequent updates, elimination and replacement. And regulation of handling waste should be tightened."

More Articles from SCMP

26 Hong Kong pupils fall ill during study tour to Guangdong cities

Chinese officials told to stop avoiding media questions about emergencies

Chinese smartphone giant Oppo, PolyU to launch joint AI research centre in Hong Kong

Chinese scientists say they have made converged energy beam weapon a reality

This article originally appeared on the South China Morning Post (www.scmp.com), the leading news media reporting on China and Asia.

Copyright (c) 2024. South China Morning Post Publishers Ltd. All rights reserved.