A new Basel Action Network (BAN) report warns that the waste produced by AI data centers has been seriously underreported, and the problem is growing. The e-waste from AI alone could amount to around 23 million shipping containers by 2050, which would be roughly enough 40-foot containers to circle the world six times laid end to end.
BAN points out that the report, released today, takes into account all the infrastructure required for servers inside data centers. This expanded view reveals more clearly the physical waste that AI produces, according to the nonprofit. Jim Puckett, the group’s founder and chief of strategic direction, argues that the apparent weightlessness of AI conceals the hardware beneath it.
“AI may feel weightless, but every model depends on an enormous amount of highly specialized, cutting edge hardware,” Puckett says 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.”
The Scale of the Problem
Only about a quarter of the 68.3 million tons of e-waste created each year worldwide is formally collected and recycled. Most of the rest goes into shadowy “informal” waste collection, where burning or burying equipment exposes workers and the local environment to toxic materials like lead and chromium.
America is the country with the most data centers, yet it has not ratified the Basel Convention, which governs the international trade of hazardous waste. American recyclers continue to export e-waste overseas, and investigations show that it frequently ends up at “backyard recycling.” The World Health Organization says these young workers and residents face significant health risks from the informal recycling there.
By 2050, BAN projects global electronic waste could reach up to 211 million metric tons per year, more than tripling its current level. Roughly 15 to 20 percent of that total is blamed on artificial intelligence. The group’s estimate goes beyond earlier AI waste forecasts centered on servers and GPUs, now covering power supply and distribution, cooling systems, backup power, and networking equipment.
BAN also counts what it calls “AI Waste Contagion” — a wide-ranging category that incorporates everything from telecommunications infrastructure to personal devices likely to become obsolete and replaced earlier as a result of advances in AI.
Comparing Previous Estimates
Previous estimates of AI’s e-waste have been more conservative. But by focusing on servers and accelerators, past studies miss about 87 percent of a data center’s electro-mechanical infrastructure, according to BAN.
| Estimate | Scope | Result |
|---|---|---|
| BAN’s new report | Servers, GPUs, power supply, cooling, backup, networking, AI Waste Contagion | Up to 20 percent of global e-waste by 2050 |
| 2024 study | AI e-waste | 1.2 million to 5 million tons by 2030 |
| February study | AI servers | 131,000 to 225,000 tons annually by 2030 |
A February study estimated that AI servers might generate anywhere from 131,000 to 225,000 tons of electronic waste per year by the end of the decade, a figure that remains roughly equivalent to the total e-waste output of a nation the size of Denmark.
What the Numbers Add Up To
The BAN projection is for 70,000 metric tons of e-waste per gigawatt of data center capacity. The figure is based on a McKinsey forecast that puts total data center capacity at up to 219GW by 2030.
BAN’s estimate for AI-related electronic equipment retired between 2025 and 2050 is between 395 to 617 million metric tons of e-waste by mid-century. The annual rate works out to roughly 8.6 million to 13.1 million metric tons of equipment retired each year due to AI.
The gap between the two numbers is large. A February study limited its scope to servers alone and arrived at a smaller yearly total. By contrast, BAN’s broader tally places AI waste at a far larger share of the expected e-waste surge.
Puckett spells out the danger plainly: the expansion could turn into a disaster more severe than the current one. The data show the difference between what gets made and what gets watched is widening quickly.
The argument at the heart of the report is straightforward: the hardware that runs AI is specialized and accumulates fast. It piles up when it fails. The size of that accumulation is now being measured in shipping containers moving around the world.
It has yet to be determined how companies and governments will react to this new wave of waste. What is clear is that the waste exists, that its true extent has been underestimated, and that its volume continues to grow.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.