AI e-waste volumes are significantly underestimated

A report from the Basel Action Network (BAN) suggests that the volume of electronic waste generated by artificial intelligence is significantly underestimated. Current estimates often focus exclusively on servers and accelerators, which represent only about 13% of a data center’s electromechanical infrastructure.

The study identifies that the remaining 87% of infrastructure—including networking, power distribution, storage, backup, and cooling systems—is also classified as e-waste. This uncounted equipment totals approximately 70,000 metric tonnes per gigawatt of capacity.

Furthermore, the report notes that the rapid turnover of GPUs and the industry’s operational practices are compressing equipment lifespans to between 2.5 and 5 years. Consequently, the study projects that by 2030, the amount of AI-driven electronic equipment being retired could be 40 to 60 times higher than previous academic projections.