The rapid advancement of digital technologies has led to the proliferation of electronic devices and systems, resulting in an alarming increase in electronic waste (e-waste). GenAI, in particular, requires substantial computational resources for model training and inference, but the impact of this on e-waste is not fully understood.
The latest Global E-Waste Monitor by the United States Institution for Training and Research (UNITAR) reveals that the world’s generation of e-waste is rising five times faster than documented e-waste recycling.
In 2022, the world produced 62 million tonnes of electronic waste, marking an 82% rise from 2010 levels. If this trend persists, global e-waste generation is projected to increase by an additional 32%, potentially reaching 82 million tonnes by 2030.
Why does it matter? E-waste, such as discarded electronic equipment and products, is a major threat to public health and safety. Some of the e-waste such as batteries contain toxic additives and hazardous substances that can cause serious harm to human health.
When consumers dispose of their electronics improperly or fail to recycle, the cycle of demand for new materials, like cobalt, continues to grow.
Cobalt is a critical component used in many electronic devices, including batteries. The world’s largest cobalt reserve is located in the Democratic Republic of the Congo (DRC), where over 255,000 citizens, including 40,000 children, mine cobalt under hazardous conditions. This relentless need for more cobalt exposes these workers to dangerous working conditions, toxic exposure, and long-term health risks.
While the threat of e-waste is not new, GenAI is making it significantly worse. The main source of e-waste generated by Generative AI comes from the high-performance computing equipment used in data centers and server farms, such as servers, GPUs, CPUs, memory units, and storage drives.
Some e-waste contains valuable metals such as gold, silver, copper, and rare earth elements. However, it also contains hazardous materials such as chromium, lead, and mercury.
According to an article published in the New York Times, “backyard recyclers” in Thailand, India, and Indonesia use nitric and hydrochloric acid to wash discarded circuit boards to recover gold. They are literally cooking the e-waste to get valuable metals hidden inside plastic, circuits, and wires.
A key factor contributing to the high levels of e-waste generated by AI companies is the rapid pace of technological advancement in hardware. The typical lifespan of high-performance computing devices is just two to five years, after which they are replaced by more advanced models.
Shutterstock/Ozrimoz
While several countries have enacted e-waste legislation, the approaches vary widely, leading to inconsistent and fragmented solutions. There is currently no federal law in the US that supports e-recycling. The insufficient recycling infrastructure and a lack of incentives for manufacturers to improve product design for recyclability have further exacerbated the issue.
In the paper “E-Waste Challenges Of Generative Artificial Intelligence” published in Nature, the authors shared that expanding the lifespan of technologies is one of the most significant ways to cut down on e-waste.
Only about 22% of e-waste is recycled today. The researchers recommend reusing or refurbishing components to cut down on e-waste. It would also help to design hardware in a way that makes it easier to recycle and upgrade. The study predicts that these strategies can reduce e-waste by up to 86% in the best-case scenario.
As the issue of e-waste receives growing attention, there is increasing pressure on tech companies to make it easier for consumers to recycle and repair electronic devices. However, an increase in the lifespan of the hardware may not be in the best interests of these companies. As a result, the responsibility may shift to end users to decide if they want to prioritize sustainability.
