The ongoing campaign to build thousands of massive datacenters for AI will create a “tsunami” of discarded electronics, putting the world on a course to triple its annual generation of toxic e-waste in the next 25 years, a new report projects.

The electronic trash AI generates in the next 25 years could fill enough shipping containers to encircle the globe six times, it said. Yet, says the report by the non-profit environmental group Basel Action Network (BAN), there is no plan to deal with that waste – which typically contains toxics like lead, mercury, cadmium and forever chemicals, as well as valuable metals that are challenging to reclaim.

“Already electronic waste is the fastest-growing waste stream on the planet and only one-fifth is properly managed,” said Jim Puckett, co-founder of BAN, which works to see that toxic waste doesn’t end up getting dumped on the world’s poorer communities.

Experts concur. “We’re doing a terrible job of handling e-waste in the US,” said Vince Beiser, author of the book Power Metals, which details the global problems of trying to reclaim valuable metals inside electronic goods. “Some of it gets dumped overseas. Most of it just gets dumped in landfills, which is a problem because it can leach all kinds of nasty things into the water, into the soil.”

There are 4,871 new datacenters in the construction pipeline in the US, the report said, adding that, while not all are clearly designated for use by AI, many will be.

It calculated the volumes of e-waste related to computer servers used for AI, and to “the thousands of tonnes of copper, steel, concrete” for cooling, power distribution, networking and energy storage equipment needed to operate inside datacenters.

To estimate the total, the report estimated the amount of computing power analysts project will be built – totalling roughly 100 gigawatts of new capacity worldwide by 2030 alone and growing beyond that. It then analyzed how much equipment would be needed by weight – including racks of computer servers weighing 1,360kgs each, switches that weigh up to 30kgs, and hundreds of kilograms of copper cable connecting each rack.

Such equipment won’t be designed for repair and reuse, the report projects, and would need to be completely replaced every two to five years.

On top of this, it assumed that AI developments will force the rapid replacement of consumer computers, phones, and telecommunications equipment owing to the computing requirements of the new systems.

By taking this wide view of all the various types of e-waste that will be created, the BAN report projects that AI-driven equipment retirement will generate up to 13m metric tonnes per year by 2030 – a far higher estimate than a handful of previous studies.

AI datacenter construction has already created a groundswell of bipartisan opposition around the US for the huge amount of space and energy it requires, but the report comes when focus is turning to the long-term pollution AI could create.

A group of environmental and health experts last week called for the federal government to issue a pledge to protect citizens’ health from the repercussions of AI buildout. It cited estimates that the air pollution from the enormous amount of electricity generation needed for datacenters could result in hundreds of thousands of new asthma cases and as many as 1,300 premature deaths per year in the US alone.

The Trump administration has sought to weaken environmental safeguards that would normally govern such a massive technological buildout.

A 2025 report by McKinsey & Company estimates that, in order to achieve the tech industry’s planned rollout of artificial intelligence, nearly $7tn in capital outlays will be needed worldwide in the next five years to build datacenters running millions of computer servers 24 hours a day.

BAN’s white paper notes that this unprecedented spending “is roughly the cost of ending world hunger for the next 75 years – being spent in just five”.

“We all find ourselves unwitting participants in the AI horserace,” said the report, “as IT users everywhere find themselves using AI whether they would like to or not, and yet almost always, without being aware of the greater consequences of doing so.”