The Invisible Trillion · The Mining and E-Waste Question, Honestly

The Summary, Honestly

The mining and e-waste problems with modern technology are real, they are worse than the energy and water problems, and they are in a lot of cases as bad as the most critical coverage suggests.

The specific problems — child cobalt mining in the DRC, informal e-waste processing in West Africa and South Asia, the environmental costs of rare earth processing — are things you should know about, and you should not let yourself be talked out of caring about them by a smooth recitation of progress statistics. The progress is real, but it has been slow, it has lagged the industry’s knowledge of the problems, and in some cases it hasn’t happened at all.

But notice where the worst of it lives, because the location matters for a book about AI. The graphic human horrors — the children in the pits, the people burning circuit boards over open fires — are mostly in the device and battery supply chain: the phones, the laptops, the electric cars. The chip itself, the fingernail-sized piece of silicon, has a modest material footprint. A data center full of GPUs is not where the cobalt and the acid baths are.

What the data center does have is a hardware life of its own — and this is the part of the story that is genuinely about the AI buildout. It runs an enormous quantity of expensive hardware extremely hot, retires it in a few years, and generates, as a result, a large and fast-growing stream of electronic waste that barely existed a decade ago. The relatively good news is that this stream is, for now, handled far better than consumer e-waste: the hardware is valuable, there is a real resale market for it, and the big operators recover most of it. The open question is whether that discipline survives as the volumes explode and the careless players take a larger share.

So there are two honest conclusions, and you should carry both. First: the worst suffering in the supply chain is much more directly about the phone in your hand than about the ChatGPT query you sent with it. Second: the buildout still creates a real, specific, growing materials-and-waste footprint of its own, separate from the energy and water costs of the last two chapters, and a book that is honest about the costs has to count it.

You can hold both truths at the same time. You should, in fact. Either one alone is incomplete.

A note on sources: the cobalt figures are from Amnesty International’s reports on the DRC, from the OECD’s due diligence guidance, and from academic studies of artisanal mining. The lithium figures are from the USGS and from academic hydrogeology studies of the Atacama. The rare earth figures are from Chinese government disclosures (such as they are) and from Adamas Intelligence. The consumer e-waste figures are from the UN Global E-waste Monitor, the most reliable public source on the topic. The data center hardware figures draw on the UN Global E-waste Monitor 2024, the Nature Computational Science study projecting AI-driven e-waste growth, reporting on hyperscaler GPU depreciation schedules and the debate over their realism, and disclosed corporate IT-asset-disposition figures. Where the numbers are uncertain — particularly on artisanal mining, on informal e-waste, and on the real working life of data center GPUs — I have said so.

All 19 chapters of The Invisible Trillion · Library · All courses