The Growing Challenge of E-Waste in AI Data Centers

E-waste generated by the AI surge is being significantly underestimated, according to a new report. By 2050, it could accumulate to enough waste to fill 23 million shipping containers—enough 40-foot containers to wrap around the globe six times if lined up.

This estimate is much higher than previous assessments because the report’s authors consider all the infrastructure necessary to support servers in data centers. This broader perspective offers a clearer understanding of the electronic waste that AI generates, as highlighted by the nonprofit Basel Action Network (BAN), which released the findings.

“AI may seem intangible, but every model relies on a vast array of specialized, cutting-edge hardware,” says Jim Puckett, founder and chief of strategic direction at BAN. “If companies and governments don’t start preparing for this impending waste wave, the current AI expansion could lead to an even more catastrophic toxic waste crisis than we’re already facing.”

Currently, less than a quarter of the 68.3 million tons of e-waste produced globally each year is properly collected and recycled. The majority ends up in informal waste collection systems, where burning or burying equipment exposes workers and the environment to hazardous materials like lead and chromium.

The U.S., which has more data centers than any other nation, has yet to ratify the Basel Convention aimed at curbing the international trade of hazardous waste. Investigations have shown that U.S. recyclers still send e-waste overseas, where it often ends up in “backyard recycling.” This situation poses health risks to millions of children who are forced to work or live near these informal recycling operations, according to the World Health Organization.

BAN forecasts that global e-waste will triple to as much as 211 million metric tons annually by 2050, with AI contributing about 15 to 20 percent of that total. Unlike earlier estimates that focused solely on servers and GPUs, BAN’s analysis includes a wider range of equipment: power supply and distribution, cooling systems, backup power, and networking gear. Additionally, BAN introduces the concept of “AI Waste Contagion,” which encompasses everything from telecommunications infrastructure to personal devices that may become obsolete sooner due to AI advancements.

In total, BAN anticipates 70,000 metric tons of e-waste for every gigawatt of data center capacity, referencing a McKinsey projection that total data center capacity could reach up to 219GW by 2030.

Considering AI-related electronic equipment that will be retired between 2025 and 2050, BAN estimates this will result in between 395 to 617 million metric tons of e-waste by mid-century, translating to approximately 8.6 million to 13.1 million metric tons of equipment retired each year due to AI.

Previous estimates of AI’s e-waste have been more conservative. However, by concentrating on servers and accelerators, earlier studies overlook about 87 percent of a data center’s electro-mechanical infrastructure, according to BAN.

A separate study published in 2024 projected that e-waste from AI could reach between 1.2 million and 5 million tons by 2030. Another study from February indicated that AI servers could generate between 131,000 and 225,000 tons of e-waste annually by the decade’s end—an amount comparable to the total e-waste produced by a country the size of Denmark.