Recently, researchers from the University of Cambridge and the Chinese Academy of Sciences published a highly anticipated paper in the journal Nature, predicting that by 2030, the rapid development of generative artificial intelligence could lead industries to produce electronic waste equivalent to over 1 billion iPhones annually. The researchers stated that their aim is to understand the practical consequences of this rapidly expanding technology in advance, rather than to restrict its use.
Image Source Note: The image was generated by AI, authorized by service provider Midjourney
In the paper, the research team mentioned that while energy consumption has long been a focal point of attention, the physical materials and discarded electronic devices associated with this process have not received adequate attention. Their research does not aim to accurately predict the number of AI servers and the resulting electronic waste, but rather to provide a preliminary rough estimate to highlight the scale of future challenges and explore potential circular economy solutions.
The researchers used different growth scenario models, including low, medium, and high growth patterns, to analyze the required computational resources and their lifespan. The results indicate that from 2600 tons of electronic waste in 2023, the waste amount could grow to between 400,000 and 2.5 million tons by 2030, with a potential increase of up to a thousand times.
It should be noted that the 2600 tons figure for 2023 might be slightly misleading, as many computing infrastructures have been deployed in the past two years and have not yet been counted as waste. However, this figure can indeed serve as a reference standard for changes in electronic waste before and after the arrival of the generative AI wave.
The researchers proposed several possible methods to mitigate the growth of electronic waste, such as downgrading servers after their useful life instead of discarding them, or reusing their communication and power components. Additionally, improvements in software and efficiency can extend the effective use time of specific chips or GPUs. The study mentioned that rapid updates to the latest chips could be beneficial, as failing to upgrade in time might require businesses to purchase two lower-performance GPUs to accomplish the work of a single high-end GPU, thereby exacerbating the generation of electronic waste.
By implementing these mitigation measures, the researchers estimate that the amount of electronic waste produced could be reduced by 16% to 86%. However, whether this reduction can be achieved depends more on whether these measures will be adopted and the strength of their implementation. If every H100 chip can continue to be used in low-cost inference servers at universities, the future pressure of electronic waste will be significantly reduced; conversely, if only one-tenth of the chips are reused, the electronic waste issue will remain severe.
Key Points:
🌍 By 2030, generative AI could produce more than 1 billion iPhones' worth of electronic waste annually.
♻️ Researchers suggest reducing electronic waste through downgrading and reusing components.
📊 The amount of electronic waste can be reduced by 16% to 86%, depending on the adoption and implementation of these measures.