The Invisible Trillion · What Chips Actually Do
Why This Matters for Everything Else
Here’s why I wanted to make the point that chips are heterogeneous.
When someone says “chip manufacturing is environmentally bad,” they might mean: “the leading-edge GPU fabs use a lot of water and power.” Which is true. Or they might mean: “the global chip industry is bad for the planet.” Which is a much bigger, vaguer, and largely wrong claim, because the vast majority of chips — the microcontrollers, the sensor chips, the older-process chips in your microwave — have a modest environmental footprint and enable things like medical devices and energy efficiency that have big environmental benefits.
When someone says “AI is using all the chips,” they might mean: “the most advanced GPUs are in short supply and expensive.” Which is true. Or they might mean: “AI is consuming the world’s chip production.” Which is false — AI uses a small and fast-growing fraction of chip production, overwhelmingly concentrated in one specific kind of chip (leading-edge GPUs) from one specific manufacturer (TSMC) made for a handful of specific buyers (Nvidia and a few custom-chip companies). The microcontrollers in your dishwasher are not in shortage because of AI.
When someone says “we need to reduce chip dependence,” they might mean: “we need a backup source for leading-edge logic chips in case of a Taiwan crisis.” Which is a sensible strategic concern. Or they might mean: “we should make fewer chips.” Which would mean: fewer medical devices, fewer safety sensors in cars, fewer grid sensors managing renewable energy, fewer satellites, fewer of a lot of things that most people wouldn’t trade away even if they thought they would.
Most discourse about chips is vague in ways that conceal these distinctions. Keeping them straight is, again, most of what good thinking about this stuff consists of.
What Ends Up Where
One last picture to leave you with before the next chapter.
The chips in your phone and laptop? They live with you, in your hand or on your desk. If you get a new phone, the chips in the old one will eventually end up in recycling or landfill (see Chapter 7).
The chips in your car, your microwave, your thermostat, your medical devices? They live in those devices, sometimes for 20 years.
The CPUs, GPUs, and memory chips that handle the cloud side of things — the part of the internet that isn’t in your hand — live somewhere else entirely. They live in enormous buildings, mostly in a few dozen places in the world, running 24 hours a day, every day of the year. The websites you visit, the games you play online, the AI models you use, the streams you watch — all of them are served from these buildings.
Those buildings are what the next chapter is about.
They’re strange places, and almost no one has seen one.
A note on sources: the phone chip counts are from public teardown analyses by iFixit, TechInsights, and System Plus. The Nvidia/AI GPU history draws on Chris Miller’s Chip War and on Nvidia’s own public filings. The microcontroller and specialized-chip figures are from the World Semiconductor Trade Statistics (WSTS), the industry’s standard aggregator. When I’ve given specific chip counts for specific devices, I’ve taken them from publicly-available teardowns; they will drift over time as designs evolve.
Questions to talk about
No answer key. These are for arguing over, with someone who read the same chapter.
- Thirty to sixty separate chips in one phone, not one. How far off was your guess, and what does the gap tell you about how we picture technology?
- 'A chip' is not a thing — the word covers radically different devices at radically different scales. Where else does one word hide a whole universe?
- When people say 'AI chips' they mean one small corner of a large varied world. Why does the small corner get all the attention?
All 19 chapters of The Invisible Trillion · Library · All courses