Generative artificial intelligence (GAI) is rapidly transforming our world. Groundbreaking advancements are coming through thick and fast in art, scientific discovery and everything in between.
However, the dizzying speed of GAI development belies a troubling afterthought. A new study from the Chinese Academy of Sciences reveals a pending surge of electronic waste (e-waste) generated by the immense computational power required to train these sophisticated models. This is on top of the mountain of e-waste that already poses an existential threat to our water sources and finite materials.
The Global Environmental E-Waste Problem: What Can You Do About It?
E-waste is the world’s fastest growing waste mountain. But there are ways to stop this development. Read more here.
Enough e-waste to circle the earth—and then some
The research, conducted via a novel Computational Power-driven Material Flow Analysis (CP-MFA) model, focuses specifically on large language models (LLMs). These models—the engines behind many popular AI applications—demand vast quantities of computational hardware for data processing and iterative training. This hardware has an extremely limited lifespan. It’s not uncommon for data centre Graphics Processing Units (GPUs) used for AI to last for just a couple of years. This is especially when used at higher workloads, such as when training GAI.
The numbers are concerning, to say the least. Even under a conservative scenario, cumulative e-waste from LLMs alone is estimated to reach 9 million tons globally by 2030. To put that into perspective, the total volume of e-waste produced in 2022—just before the widespread proliferation of LLMs—was 62 million tons. This would fill 1.55 million 40-tonne trucks, roughly enough trucks to form a bumper-to-bumper line encircling the equator, according to a report from ITU and UNITAR. In a scenario where LLM proliferation accelerates toward the upper end of expectations, the volume of LLM e-waste could soar to a staggering 16.1 million additional tons.
GAI: A ticking time bomb of toxic tech
The very nature of e-waste makes it complicated, costly and dangerous to recycle. This type of waste contains substantial quantities of both valuable recyclable materials and hazardous substances, both of which pose a complex challenge for waste management and resource recovery systems. E-waste recycling infrastructure, as it stands, is nowhere near ready to cope with the deluge we’ve already created. According to a recent report by the UN, the rise of e-waste is five times faster than e-waste recycling rates can keep up with. To put it another way, less than a quarter of e-waste mass was documented as having been properly collected and recycled in 2022. Potentially adding another 25 percent more e-waste to this existing avalanche will put untold pressures on our already ill-equipped infrastructure.
The impact of e-waste
E-waste is full of toxic metals like lead, mercury, cadmium and beryllium, which, when discarded in landfills, pose a major threat to soil and water.
When burned, e-waste releases poisonous compounds into the air, which can then seep into groundwater, endangering both aquatic and land-based life.
The World Health Organisation warns that exposure to e-waste is linked to several severe health problems, including respiratory issues, altered lung function and adverse birth outcomes. And, as the volume of e-waste has grown, so have these health challenges.
The regional distribution of this e-waste also warrants attention. The Chinese Academy of Sciences study identifies North America and East Asia as the primary contributors to LLM-related EoS waste. Whilst unsurprising, this foregrounds the geographical concentration of AI’s development. But, what about its post-use treatment? You guessed it. The Global South.
Pathways to progress could already exist
But, there are potential pathways towards mitigating this growing GAI e-waste challenge. The study suggests that implementing circular economy strategies throughout the lifecycle of AI server components—from extending their lifespan and promoting reuse to optimising material recovery and responsible recycling—will be crucial. Designing more energy-efficient and hardware-lean AI algorithms could also play a significant role in reducing the overall material demand. Work is already being done in this area, although the results are, so far, yet to be adequately measured.
The capabilities and applications of Generative AI will only continue to expand, demanding ever more resources and generating an alarming surge of e-waste. While the focus often lies on AI’s energy consumption and carbon footprint, the material dimension needs urgent scrutiny. The dazzling promise of AI must not blind us to its very real physical footprint. As we stand on the cusp of an AI-driven future, the responsibility lies with AI developers, policymakers and governments to champion solutions which serve humanity sustainably instead of leaving behind an irreparable legacy of waste.
