The Basel Action Network (BAN), a non-profit organization, has released a report predicting that retired AI-related electronic devices will cumulatively generate between 395 and 617 million tons of e-waste from 2025 to 2050. According to their estimates, this waste could fill approximately 23 million 40-foot containers.
BAN estimates that by mid-century, AI-related devices will generate approximately 8.6 to 13.1 million tons of e-waste annually. The organization also predicts that total global e-waste could increase to 211 million tons per year by 2050, with about 15% to 20% related to AI.
This estimate adopts a broader statistical scope than previous studies. In addition to servers and GPUs, the report includes power distribution equipment, cooling systems, backup power supplies, and network devices. BAN believes that counting only servers and accelerators would omit approximately 87% of the electromechanical infrastructure in data centers.
The report also introduces the concept of "AI waste proliferation," which accounts for telecommunications infrastructure and personal devices that may be prematurely obsolete due to advances in AI. According to BAN's calculations, each gigawatt of data center capacity corresponds to approximately 70,000 tons of e-waste; the report also cites McKinsey's forecast that global data center capacity could reach up to 219 gigawatts by 2030.
Jim Puckett, founder and Strategic Platform Director at BAN, stated that each model relies on large amounts of highly specialized advanced hardware, and businesses and governments need to plan ahead for the disposal of related waste. The report points out that currently, the world generates 68.3 million tons of e-waste annually, with less than a quarter being formally collected and recycled.
Devices not entering formal recycling systems may be incinerated or buried, exposing workers and local environments to toxic substances such as lead and chromium. In comparison, a 2024 study predicted that AI e-waste could reach 1.2 to 5 million tons by 2030; another study estimated that AI servers could generate 131,000 to 225,000 tons of e-waste annually by the late 2020s.