AI e-waste is projected to be vastly higher than prior estimates, with BAN’s Part 1 estimating 8.6–13.1 million tonnes retired annually by 2030 and 31–46 million tonnes by 2050 when accounting for networking, power, storage, cooling, and early scrap beyond servers and GPUs.

The report introduces a category called “AI waste proliferation,” which includes telecommunications infrastructure and consumer devices that may become obsolete due to AI development.

BAN estimates that each gigawatt of data-center capacity could generate about 70,000 metric tonnes of electronic waste, with a McKinsey forecast suggesting data-center capacity could reach 219 GW by 2030.

The United States has not ratified the Basel Convention to restrict hazardous-waste trade and has been shown to ship e-waste abroad for informal processing, raising health concerns for workers and nearby communities.

The article references The Verge and UA.News Telegram channel for additional details and updates.

A significant portion of current e-waste is inadequately recycled, with much of it exported to polluting operations in developing countries; BAN calls for immediate budgeting, infrastructure, and policy measures to mitigate this looming toxic waste wave.

Global e-waste generation stands at about 68.3 million tonnes annually, with less than a quarter formally collected and recycled.

Current e-waste practices are insufficient: only about a quarter of global 68.3 million tonnes is formally collected and recycled, and a large portion ends up in informal recycling that risks toxic exposure.

The report notes potential “rin-and-replace” transitions driven by technologies like superconducting power, liquid cooling, and new battery chemistries that could accelerate hardware retirement beyond planned schedules.

If the upper estimate of 617 million tonnes were realized, it would equal about 23 million 40-foot shipping containers, which could encircle the Earth six times in a line.

Other studies have produced lower range estimates for AI-related e-waste, illustrating divergence in methodologies and BAN’s broader scope in its assessment.

BAN’s calculation includes all data-center electromechanical infrastructure beyond servers and GPUs—power supplies, cooling, backup power, and networking—suggesting roughly 87% of a data center’s infrastructure is not accounted for in prior estimates.