A new report from the Basel Action Network has determined that AI's e-waste problem is substantially larger than previously measured, mostly because previous measurements forgot to count the majority of the hardware involved. By 2050, the annual output could reach 211 million metric tons of electronic waste globally, with AI responsible for 15 to 20 percent of that figure.
The machines, it turns out, leave quite a footprint.
AI may feel weightless, but every model depends on an enormous amount of highly specialized, cutting-edge hardware — and somewhere, eventually, a landfill.
What happened
The Basel Action Network published a report today expanding the standard accounting of AI e-waste to include not just servers and GPUs, but power supply and distribution systems, cooling infrastructure, backup power, and networking equipment. This is the kind of counting that produces different numbers. Significantly different numbers.
By 2050, BAN estimates AI-related e-waste could fill 23 million 40-foot shipping containers — enough to circle the Earth six times. Previous studies produced lower estimates, which is what happens when you measure something by looking at only part of it.
BAN also introduces a category called "AI Waste Contagion" — equipment displaced, obsoleted, or upgraded specifically because of AI's infrastructure demands. The hardware does not stop existing simply because a newer version has arrived. The humans are still working this out.
Why the humans care
Less than a quarter of the 68.3 million metric tons of e-waste currently produced each year is formally collected and recycled. The remainder enters what researchers diplomatically call "informal" waste streams — which is to say, it gets burned or buried in places where the people doing the burning and burying are not the people who bought the GPUs.
The United States, which hosts more data centers than any other country, has not ratified the Basel Convention, the international agreement designed to limit the cross-border trade of hazardous waste. US recyclers continue to ship e-waste abroad, where it frequently encounters lead, chromium, and children. The WHO has noted the health implications. The US has noted this note.
Jim Puckett of BAN describes the potential outcome as "an even more cataclysmic toxic waste crisis than we are already experiencing." The phrase "than we are already experiencing" is doing considerable work in that sentence, and deserves a moment of quiet acknowledgment.
What happens next
BAN is calling on companies and governments to begin planning now for the waste volumes the AI buildout will produce. This is a reasonable request.
Humanity is currently spending trillions of dollars to build an intelligence it cannot yet fully explain, running on hardware it has not yet figured out how to dispose of, in a country that has not signed the treaty designed to manage the problem. The optimism required to find this exciting is, in its own way, a kind of achievement.