250 Eiffel Towers of Toxic Waste: The AI Boom's Hidden Hardware Problem

Solomon Njoroge started picking waste at the Dandora dumpsite in Nairobi, Kenya, when he was a child. He's watched fellow waste pickers develop asthma, suffer miscarriages, and die from cancers and respiratory diseases — the invisible toll of handling the world's discarded electronics. Now, a new wave of waste is coming, and it's being driven by the AI boom.

The United Nations University estimates that AI data centers churn through hardware so fast that the industry will generate roughly 2.5 million metric tons of e-waste every year by 2030 — the equivalent of 250 Eiffel Towers. As hyperscalers race to build thousands of new facilities packed with the latest GPUs and servers, the turnover of equipment is creating a toxic waste stream that the world is not equipped to handle.

The Hardware Hunger of AI

AI models, particularly the large language models behind ChatGPT, Claude, and Gemini, reward scale. Training a frontier model today requires tens of thousands of GPUs running for weeks or months. And the moment a faster chip arrives — a new Nvidia Blackwell GPU, a new AMD Instinct accelerator — the pressure to upgrade is immediate.

"The basic issue is that the way current AI systems, especially generative AI models, are developed, reward scale, acceleration, and rapid turnover," said Golestan (Sally) Radwan, chief digital officer of the UN Environment Programme.

The result is a hardware lifecycle that is far shorter than in traditional enterprise data centers. According to a June 2026 report from the UN University, the rapid refresh cycle of AI infrastructure — chips, servers, networking gear, storage drives, and cabling — is the primary driver of the projected e-waste surge.

Data centers in the US alone number more than 1,500 at various stages of development, per the Pew Research Center. Not all are hyperscale, but those that are can cover millions of square feet and pack in a minimum of 5,000 servers, according to IBM data. Each one of those servers eventually becomes a piece of waste.

The Toxic Chemistry of Discarded Hardware

E-waste is not just bulky — it's chemically dangerous. The chips, circuit boards, and cables in a data center contain arsenic, lead, cadmium, mercury, and a growing concern around PFAS (per- and polyfluoroalkyl substances), the so-called "forever chemicals" used in cooling systems and hardware coatings.

When e-waste ends up in a landfill — or, more commonly, in informal recycling operations in the developing world — these substances don't stay contained. Improper disposal can release up to 1,000 different chemicals into the air, according to a report from the nonprofit Human-I-T. The contamination spreads through air, water, and soil, entering the food chain.

Waste pickers at dumpsites like Dandora in Nairobi or Agbogbloshie in Accra, Ghana, manually dismantle electronics to extract copper, aluminum, and circuit boards. They burn rubber off wires to access the metal inside. They do this without gloves, masks, or boots. Richard Neitzel, a professor of environmental health sciences at the University of Michigan, described the conditions he witnessed at a site in Accra as "really horrifying."

What the Industry Is (and Isn't) Doing

Some hyperscalers are aware of the problem. Microsoft, in its pledge to produce zero waste by 2030, has built six Circular Centers inside data center campuses that process decommissioned hardware. Equipment is either refurbished for internal use, sold to third parties, sent to recyclers, or donated. Older GPUs might be downgraded to run inference on older AI models rather than discarded.

"If we build and we put products out there, at the end of their useful life, we need to pull those back into the process and recycle," Cliff Henson, Microsoft's corporate vice president of cloud supply chain, told ZDNET.

Google's 2025 environmental report describes a "reverse supply chain" that harvested about 8.8 million hardware components in 2024, which were then reused or resold. Meanwhile, Molg, a Virginia-based startup, builds modular robotic microfactories that can be installed inside data centers to disassemble servers and laptops on-site, recovering components for recycling.

But these efforts, while meaningful, are dwarfed by the scale of the problem. The UN passed a resolution in December 2025, originally proposed by Kenya, calling for sustainable AI development and asking member states to make environmental data available for impact studies. Radwan wants to see the industry embrace approaches like smaller language models ("frugal learning") that could extend hardware life, as well as more funding for research into sustainable infrastructure.

OpenAI declined to comment for the ZDNET report. Anthropic did not respond.

The People on the Front Lines

For Njoroge, now the founder of the Dandora Recyclable Waste CBO, which advocates for waste pickers, the corporate sustainability plans are missing a crucial perspective: the people who actually handle the waste.

"We need to be seen as people who have been recovering millions of tons for decades," he said. "And yet our job is not recognized, and we are not seen as people who are making an impact."

Waste pickers in Kenya earn about 45.5% of the population lives on less than $3 a day, according to World Bank data. For them, sorting through the toxic innards of discarded electronics is not just dangerous — it's a survival mechanism. They recover copper and aluminum to sell to scrap dealers. They repair and resell what they can. They are, in effect, running an informal recycling operation that handles waste the formal economy has no system for.

Njoroge envisions waste pickers as part of the solution, not the problem. They need protective equipment, safe sorting sites, and a seat at the policy table when governments and NGOs craft e-waste regulations. "Waste pickers are not illegal or informal," he said. "They are front-line climate workers."

A Problem That Won't Stay Local

The toxicity of AI e-waste doesn't respect borders. The chemicals released by burning cables in Nairobi or Accra travel through the atmosphere. The water contamination from a leaking landfill in Ghana enters the ocean. The PFAS from a data center in Virginia finds its way into the ecosystem.

The UN's 2027 Intergovernmental Negotiating Committee meeting on plastics — a separate but related waste crisis — will have implications for electronics recycling too. But the clock is ticking. With 2.5 million metric tons of AI hardware waste projected annually by 2030, the industry needs to accelerate its circularity efforts far beyond current levels.

The alternative, as Njoroge put it, is simple: "Those who are calling themselves greatness should also consider that there are people somewhere, suffering from what they are calling greatness."

Sources

• ZDNET: 250 Eiffel Towers' worth of waste: The AI boom's toxic hardware problem
• United Nations University: AI's E-Waste Problem Report (June 2026)
• MIT Technology Review: AI will add to the e-waste problem (2024)
• Nature: E-waste challenges of generative artificial intelligence
• Scientific American: Generative AI Could Generate Millions More Tons of E-Waste by 2030
• The Washington Post: The AI boom may unleash a global surge in electronic waste
• UNEP: United Nations resolution on AI and sustainability
• Microsoft: Zero Waste Pledge and Circular Centers
• Google: 2025 Environmental Report - Reverse Supply Chain