A new report expands the scope of AI e-waste statistics from servers and GPUs to entire data centers, multiplying the numbers by 40 to 60 times: between 2025 and 2050, cumulative AI-related electronic equipment retirement will reach 395 million to 617 million metric tons—at peak volume enough to fill about 23 million standard shipping containers, which placed end to end would circle the Earth six times. The real bulk isn't chips, but the long-overlooked electromechanical supporting infrastructure: power supply and distribution, cooling, backup power, and the rest. With formal global recycling rates below a quarter, this bill is quietly ballooning.

Past estimates of AI e-waste were like counting only the pots in a restaurant's kitchen while ignoring the stoves, exhaust ducts, and grease filters—every new dish on the menu ages the entire kitchen chain. Data centers work the same way: GPUs are the stars, but what truly determines how long they run and how much waste piles up when they're retired is that entire invisible layer of electromechanical muscle. The analogy ends there—the real difference is that swapping a restaurant pot is a minor hassle, while swapping data center electromechanical infrastructure means hundreds of millions of tons of toxic hardware flowing into the recycling chain, and formal global recycling capacity still can't cover a quarter of it.
Event

The AI e-waste tally has been off by 40 to 60 times

Expand the count from GPUs to entire data centers, and the AI e-waste estimate jumps 40 to 60 times. That is the finding behind the September 15 report from the Basel Action Network.

The Seattle-based nonprofit Basel Action Network (BAN(an international NGO that tracks toxic waste trade)) published a report posing a question: just how much e-waste does AI leave behind?

Previous research focused only on core computing hardware like servers and GPUs. To operate, a data center also needs a full suite of electromechanical supporting systems: power supply and distribution, cooling systems, backup power, and network equipment. BAN expanded the scope from chips to entire facilities, and the estimated scale jumped by 40 to 60 times.

395–617 million tons
Cumulative projected AI-related e-waste, 2025–2050
Source: Basel Action Network (BAN)
23 million
Volume if packed into 40-foot standard shipping containers
Source: Basel Action Network (BAN)
About 87%
Share of electromechanical infrastructure missed by traditional tallies
Source: Basel Action Network (BAN)
Below 25%
Share of global e-waste formally collected and recycled
Source: Basel Action Network (BAN)

How did they arrive at these figures? The report cites a McKinsey projection that global data center capacity will peak at 219 GW(gigawatt—a unit for measuring power plant or data center capacity) by 2030. BAN then applied a conversion factor of 70,000 tons of e-waste per 1 GW of capacity—and electromechanical supporting systems account for the lion's share.

Why It Matters

What does it mean when the official tally misses 87% of the hardware?

The hardware cost of AI has long been undercounted. Past tallies only tracked servers and GPUs; the real bulk is in the electromechanical supporting infrastructure.

That gap is the story. BAN added power supply, distribution, cooling, backup power, and network equipment to the count and found this supporting infrastructure makes up roughly 87% of data center weight—weight that never showed up in earlier estimates. By BAN's calculation, every 1 GW of data center capacity corresponds to 70,000 tons of e-waste; McKinsey projects global data center capacity could peak at 219 GW by 2030.

What touches ordinary people is "AI waste sprawl": telecom infrastructure needs upgrading, and personal devices made prematurely obsolete by AI advances need replacing. The higher the computing demands, the faster the turnover, and the sooner waste enters the recycling chain—formal channels simply can't keep up.

The world already generates 68.3 million tons of e-waste each year, and less than a quarter of it is formally collected and recycled. The rest mostly enters "informal" recycling systems, where equipment is burned or landfilled, releasing toxic substances like lead and chromium. The World Health Organization has noted that millions of children working in or living near recycling sites face health risks as a result.

Now stack AI's projected growth on top of a system that can't handle today's volume. The United States hosts more data centers than any other country, yet has not ratified the Basel Convention, which restricts international trade in hazardous waste; investigations have found that U.S. recyclers are still shipping e-waste overseas, where it flows into "backyard recycling" operations.

How the mechanism works, and what you can do about it—broken down below.