Restrictions on AI chip imports and exports could add up to around 5.7 million tons of waste by the end of this decade.
The frequent and large-scale optimization of artificial intelligence (AI)-based systems threatens to worsen the pollution crisis generated by the technology industry. The emerging sector could produce electronic waste equivalent to more than 13 billion iPhone 15 Pro units by the beginning of the next decade. The projection is yet another warning of the impacts that solutions such as ChatGPT or Gemini have on the environment.
The figure is the result of research by the University of Cambridge and the Institute of Urban Environment of the Chinese Academy of Sciences. The work published in the journal Nature Computational Science explains that the accelerated advance of AI requires increasingly sophisticated computational requirements. The useful life of servers is reducing at the same speed.
The researchers have created a model that provides initial rough estimates of the waste stream associated with AI hardware . “Our work does not aim to accurately predict the amount of waste that AI servers will generate in the future. The goal is to provide rough estimates to demonstrate the magnitude of the problem ahead and explore possible solutions,” they explain.
The team took the Nvidia DGX H100 server as a reference. The computing platform is made up of eight graphics processing units (GPUs). Today, it supports most of the major next-generation digital services. The authors laid out four possible scenarios for the growth and adoption of AI:
- Limited (41% growth): takes as a reference the growth rates in the use of AI recorded between 2022 and 2023. It establishes a non-massive adoption of the new technology.
- Conservative (85%): believes that AI could adopt a gradual and sustained pace of penetration and development, similar to that of voice assistants.
- Moderate (115%): Suggests that AI’s popularity will increase rapidly and widely thanks to its integration into commonly used digital platforms such as social media.
- Aggressive (136%): assumes that large language models will become “a ubiquitous tool in people’s daily lives.” Therefore, AI would be used in a widespread, massive and constant manner.
Circular economy could reduce the impact of AI
The experiment’s findings indicate that waste generation would grow from the 2,600 tons documented last year to 2.5 million tons in 2030. The volume would be equivalent to throwing away between 2.1 and 13.3 billion iPhone 15 Pro units . The calculation predicts that no strong measures will be implemented in the next five years to reduce the waste generated by the digital industry.
The analysis adds that restrictions on semiconductor imports and exports could aggravate the situation. Several manufacturers have improved the efficiency of their chips and servers by integrating technologies that guarantee the same performance with fewer resources. Blockades such as those imposed by the United States limit the global adoption of these improvements. The situation could lead to a 14% increase in the number of obsolete AI servers and add up to around 5.7 million tons of waste by 2030.
The scientists say the AI industry urgently needs to adopt circular economy mechanisms to reduce its environmental footprint. They say reusing GPUs’ communication, memory and battery modules could reduce e-waste by more than 40%. “Implementing strategies of this nature throughout the generative AI value chain could reduce e-waste production by 16-86%,” they add.
The environmental impact of AI is still uncertain. Dozens of specialists have recognized the potential of technology to make the fight against the climate crisis more efficient . The United Nations Educational, Scientific and Cultural Organization has said that “big data, artificial intelligence and digital transformation can play an essential role in ensuring environmental sustainability and sustainable development.” Despite this, environmental advocates demand that the development of these resources take into account the inherent ecological footprint they cause. They ask companies to modify their processes to reduce their waste, emissions, water and energy consumption.
