Artificial Intelligence & Machine Learning , Next-Generation Technologies & Secure Development
AI Boom Set to Dump a Mountain of E-Waste
E-waste From Gen AI Hardware May Equal 2.5M Tons Per Year by 2030The hardware powering chatbots could increase electronic trash by a thousand times by the end of the decade, warn researchers.
See Also: Defending Your AI Future with Prisma Cloud
Waste generated by physical materials supporting generative artificial intelligence development and operations could equal more than 10 billion iPhones per year, showed a study from researchers at Cambridge University and the Chinese Academy of Sciences.
The study does not precisely forecast the e-waste from AI servers and related equipment, such as GPUs, CPUs, storage, internet communication modules and power systems. Instead, it works as a guideline for the industry to curb the adverse impact of the technology's rapid expansion. The study did not consider ancillary machinery such as cooling and communication units.
The researchers studied the computing requirements and the products' lifelines in scenarios of low, medium and high AI growth. Computing devices typically have two- to five-year lifespans after which they're replaced with up-to-date versions.
Projecting from current growth rates, e-waste could increase between 3% and 12% by the end of this decade. "Our results indicate potential for rapid growth of e-waste from 2.6 thousand tons per year in 2023 to around 0.4-2.5 million tons in 2030," researchers said. The study uses 2023 as its starting point, calculating e-waste before and after the public launch of ChatGPT.
Geopolitical restrictions on semiconductor imports add to the e-waste problem since the same products are manufactured in multiple countries when they could be limited to specific geographies for simpler management.
Gen AI's estimated e-waste is relatively a small fraction of the 60 million metric tons produced globally.
The researchers advise downcycling servers at the end of their lifespan and repurposing components such as communications and power. Using faster, high-end GPUs that can do the job of two low-end ones could also reduce e-waste due to the more expensive chips' longer lifespan and lower material profile.
A top reason enterprises dispose e-waste rather than recycling computers is the cost. E-waste can contain metals including copper, gold, silver aluminum and rare earth elements but proper handling is expensive. Data security is a concern as well - breach proofing doesn't get better than destroying equipment.
The researchers nonetheless estimate that taking mitigating steps could reduce e-waste between 16 and 86%. The reduction range is a reflection of projected uptake, they said. Should every GPU be reused in a low-cost inference server after it can no longer be used for AI, that act of repurposing would amount to a significant reduction. But if only one in 10 becomes repurposed, the impact diminishes. Reducing AI e-waste is a choice, not an inevitability.