AI data centre waste projected to reach 617 million tons by 2050

A new report warns that the electronic waste generated by the artificial intelligence boom has been significantly underestimated. By 2050, the accumulated trash could fill 23 million shipping containers, a volume equivalent to lining up 40-foot containers in a row to circle the globe six times. This projection is substantially higher than previous estimates because the authors have expanded the scope of their analysis to include all infrastructure required to support servers in data centres, rather than focusing solely on the computing hardware itself.

The Basel Action Network, a nonprofit organisation that published the paper, argues that this broader perspective provides a clearer picture of the physical footprint of AI. Jim Puckett, the founder and chief of strategic direction at the network, stated that while AI may appear weightless, every model relies on a vast amount of specialised hardware. He cautioned that if companies and governments do not begin planning for this surge in waste, the current buildout could result in a toxic waste crisis more severe than the one already being experienced. The report highlights that less than a quarter of the 68.3 million tons of e-waste created globally each year is formally collected and recycled. The majority is dumped into informal waste collection systems, where burning or burying equipment exposes workers and local environments to toxic materials such as lead and chromium.

The United States, which hosts more data centres than any other country, has not ratified the Basel Convention designed to limit the international trade of hazardous wastes. Investigations have found that US recyclers still ship e-waste abroad, where it often enters informal recycling processes. According to the World Health Organization, millions of children who work or live near these sites face health threats from the exposure. The Basel Action Network predicts that global e-waste will triple to up to 211 million metric tons a year by 2050, attributing 15 to 20 percent of that total to AI. This estimate includes power supply and distribution, cooling systems, backup power, and networking equipment, as well as a category termed AI Waste Contagion, which covers telecommunications infrastructure and personal devices that may become obsolete earlier due to AI advances.

The network expects 70,000 metric tons of e-waste per gigawatt of data centre capacity. Citing a McKinsey projection that total data centre capacity could reach 219 gigawatts by 2030, the report concludes that AI-related equipment retired between 2025 and 2050 will add between 395 and 617 million metric tons of waste. This equates to 8.6 million to 13.1 million metric tons of equipment retired annually. Previous studies, which focused on servers and accelerators, missed approximately 87 percent of a data centre’s electro-mechanical infrastructure. A 2024 study predicted AI e-waste could reach 1.2 million to 5 million tons by 2030, while another found that AI servers could create 131,000 to 225,000 tons annually by the end of the decade, an amount comparable to the total e-waste produced by a country the size of Denmark.