Report warns AI boom could create e-waste enough to circle Earth six times by 2050

BAN attributes the problem in part to Moore's Law. TechRadar reported that developers can no longer achieve the same year-over-year performance improvements, so they are introducing entirely new types of technology. AI companies are deploying large numbers of specialized AI accelerators alongside GPUs, cooling equipment, networking gear and other hardware, the report says.

The report also challenges earlier estimates of data center e-waste. BAN says such estimates have focused only on servers and accelerators, which represent about 13 percent of the total picture. The remaining 87 percent, spanning networking, power distribution, storage and backup, and cooling, is largely unaccounted for in most figures, according to BAN.

The scale of the buildout adds to the concern. BAN says global data center capital expenditure is expected to reach $7 trillion between 2025 and 2030, with $4.3 trillion attributed to electronic hardware. Each additional 1GW of data center capacity is estimated to require around 70,000 tonnes of electronics. With 100GW of additional capacity anticipated globally by the end of this decade, the report says the true scale of the waste stream becomes clearer.

Refresh cycles are another factor. BAN predicts that AI accelerators, servers and racks are replaced every 2.5 years, compared with the usual five to seven years for a general-purpose server. While BAN describes these as speculative estimations rather than actual quantities, it says one outcome is certain: more data centers will mean more e-waste.

To reduce some of the impact, BAN first suggests designing AI data center equipment to last longer. Recycling and second use are also major topics, but the report does not go into whether AI-specific hardware can later be used for lesser functionalities, such as being placed in regular server facilities, according to the report.

Editor's Summary

The Basel Action Network report estimates that AI-related equipment could generate 395–617 million tonnes of e-waste from 2025 to 2050, enough to fill up to 23 million shipping containers. It warns that most data center hardware is excluded from common estimates and that AI gear is replaced every 2.5 years. The report calls for longer-lasting design, recycling and reuse, though it does not assess whether AI-specific hardware can be repurposed for less demanding uses.