AI data center e-waste could reach 617 million metric tons by 2050, BAN report warns

A much bigger footprint than previously counted

The Basel Action Network (BAN), a nonprofit that tracks the international trade in hazardous waste, argues in a new report that the AI industry's electronic waste problem has been significantly understated. According to The Verge, BAN projects that AI-related equipment retired between 2025 and 2050 will add up to somewhere between 395 million and 617 million metric tons of e-waste — enough, the group says, to fill 23 million 40-foot shipping containers, or a line of containers long enough to circle the globe six times.

Why the new figure dwarfs earlier estimates

Earlier studies of AI's e-waste footprint focused almost entirely on servers and GPU accelerators. BAN contends that this framing overlooks about 87 percent of a data center's electro-mechanical infrastructure. The report widens the count to include power supply and distribution hardware, cooling systems, backup power equipment and networking gear — all of which is installed, refreshed and eventually discarded alongside the compute hardware it supports.

BAN also introduces a category it calls "AI Waste Contagion," which sweeps in equipment well beyond the data center itself: telecommunications infrastructure plus consumer devices likely to be replaced early as AI capabilities advance. That broader framing is a large part of why the new numbers tower over previous ones.

The per-gigawatt math

BAN estimates 70,000 metric tons of e-waste per gigawatt of data center capacity, and cites a McKinsey projection of global capacity reaching up to 219 gigawatts by 2030. Broken down annually, the group expects AI to drive the retirement of roughly 8.6 million to 13.1 million metric tons of equipment each year. By 2050, BAN expects total global e-waste to triple to as much as 211 million metric tons a year, with AI responsible for about 15 to 20 percent of it.

For comparison, a 2024 study put AI-related e-waste at 1.2 million to 5 million tons by 2030, and a separate study published in February estimated that AI servers alone would generate 131,000 to 225,000 tons annually by the end of the decade — an amount comparable to all the e-waste produced by a country the size of Denmark. The gap between those figures and BAN's reflects a difference in scope rather than a dispute about the servers themselves.

Where the waste actually goes

The report lands against a backdrop of weak global recycling. According to The Verge, less than a quarter of the 68.3 million tons of e-waste generated worldwide each year is formally collected and recycled. Much of the rest flows into informal channels where equipment is burned or buried, exposing workers and surrounding environments to toxic materials such as lead and chromium.

The United States, which hosts more data centers than any other country, has not ratified the Basel Convention, the treaty limiting international trade in hazardous waste. Investigations cited by The Verge have found that American recyclers continue to ship e-waste abroad, where it often slips into small-scale "backyard recycling" operations. The World Health Organization says millions of children who are forced to work in or live near such sites face serious health threats.

Why it matters

Most scrutiny of AI's environmental cost has centered on electricity and water consumption. BAN's report extends the ledger to the full hardware lifecycle, and the numbers suggest the industry is discarding infrastructure far faster than recycling systems can absorb it. Jim Puckett, BAN's founder and chief of strategic direction, warned that "every model depends on an enormous amount of highly specialized, cutting edge hardware" and that without planning by companies and governments, today's AI buildout could become an even more catastrophic toxic waste crisis than the one already underway.

For data center operators and the cloud industry at large, that points to a growing obligation — both regulatory and reputational — to plan for end-of-life handling of the physical plant behind AI, not just its deployment and power draw.