Twenty-three million shipping containers. That’s the pile of electronic junk a new report says the AI buildout will leave behind by 2050, and if you lined up 40-foot containers end to end, you’d wrap the planet about six times.
The number comes from the nonprofit Basel Action Network, which published the paper today. It lands well above every previous attempt to size this problem, and the reason isn’t a more pessimistic view of AI demand. It’s arithmetic about what a data center actually contains.
The servers were never the whole story
Earlier estimates counted servers and accelerators. BAN argues that approach misses about 87 percent of a data center’s electro-mechanical infrastructure.
So the group counted the rest: power supply and distribution, cooling systems, backup power, networking equipment. All of it wears out. All of it gets ripped out and replaced on a cycle, and none of it showed up in the older tallies.
BAN also adds a category it calls “AI Waste Contagion,” which stretches from telecommunications infrastructure to personal devices likely to become obsolete and replaced earlier as a result of advances in AI. That’s the loosest part of the model, and it’s doing real work in the total.
The math, in units you can check
BAN’s core figure is 70,000 metric tons of e-waste per gigawatt of data center capacity. It pairs that with a McKinsey projection of total data center capacity reaching up to 219GW by 2030.
Run AI-related equipment retirements from 2025 through 2050 and you get between 395 million and 617 million metric tons. That works out to roughly 8.6 million to 13.1 million metric tons retired every year.
Globally, BAN expects e-waste to triple to up to 211 million metric tons a year by 2050, with about 15 to 20 percent of it attributable to AI.
Compare that to what the earlier studies said
A study published in 2024 predicted AI e-waste could reach between 1.2 million and 5 million tons by 2030. Another, published in February, found AI servers could create between 131,000 and 225,000 tons annually by the end of the decade.
That February figure isn’t trivial on its own. It’s comparable to all the e-waste produced by a country the size of Denmark. But set it next to BAN’s annual range and the gap is roughly two orders of magnitude, which tells you the disagreement here isn’t about AI adoption curves. It’s about where you draw the box around a data center.
Why the destination matters more than the tonnage
The world already generates 68.3 million tons of e-waste a year, and less than a quarter of it is formally collected and recycled. The rest disappears into informal waste collection, where burning or burying equipment exposes workers and the surrounding environment to toxic materials like lead and chromium.
The US has more data centers than any other country. It also hasn’t ratified the Basel Convention, the treaty set up to limit international trade in hazardous waste.
Investigations have found US recyclers still ship e-waste abroad, where it often slips into “backyard recycling.” The World Health Organization says millions of children forced to work or live nearby face health threats from that informal processing.
The warning, in the report’s own words
“AI may feel weightless, but every model depends on an enormous amount of highly specialized, cutting edge hardware,” said Jim Puckett, founder and chief of strategic direction at BAN, in a press release. “If companies and governments do not begin planning for this new waste tsunami, today’s AI buildout could become an even more cataclysmic toxic waste crisis than we are already experiencing.”
Puckett’s framing is deliberately alarming, and the “Waste Contagion” category gives skeptics something to push on. But the narrower claim underneath doesn’t depend on it: the chillers, the transformers, the backup generators and the switching gear are real hardware with real replacement cycles, and until this report nobody was counting them.
If you want one number to carry out of this, make it 70,000 metric tons per gigawatt. Every time a company announces a new campus measured in gigawatts, that’s the multiplier that tells you what it leaves behind.
