Generative AI could produce 2.5 million tonnes of e-waste a year by 2030 — a thousandfold increase

The study models the server fleet needed to serve generative AI through 2030 and the hardware retired along the way. Annual e-waste rises from about 2,600 tonnes in 2023 to as much as 2.5 million tonnes a year by 2030 in the high-adoption scenario, for a cumulative 1.2–5 million tonnes. The aggressive case includes 1.5 million tonnes of printed circuit boards and 0.5 million tonnes of batteries, both containing hazardous materials.

The authors find circular-economy measures could reduce the total by 16–86%, with lifespan extension and module reuse the most effective. UC Riverside's Shaolei Ren, who was not involved, called the DGX H100-based projection a reasonable gauge.

Only two companies in the index publish hardware reuse figures: Microsoft (90.9% of retired servers and components reused or recycled in 2024, through six Circular Centers) and Google (8.8 million components harvested in 2024). ecogpt.co estimates the future e-waste of each company's fleet at 17 kg of server per accelerator — about 87,000 tonnes for Google's fleet alone.