Hardware
An electron gun on your desk
Written 27 July 2026 · published 8 September 2026
The machine that looked like serious physics was the simple one.
Most desktop computers sold before about 2005 drew their picture with a particle accelerator. Nobody said it that way at the time. It was the television-shaped thing on the desk, and mine was attached to an Atari 520STe, in 1992.
An electron gun fired a beam through a vacuum. Magnetic coils steered it across a phosphor-coated screen, redrawing the picture line by line fifty times a second. It was heavy. It got hot. It whined at a pitch adults gradually stop being able to hear, so you always knew from the doorway whether it was on.
Nothing about it was hidden. The physics was visible from outside the box, and the failures announced themselves before they had finished happening: the picture bowed, the corners went soft, degaussing thumped like something structural had given way. When one died it died in the open, with a white line and a smell.
The panel that replaced it makes no sound. It is a sheet of glass with millions of transistors printed across it, each controlling one subpixel. Thin-film transistors laid down at micron scale, over a sheet the size of a door, with a defect budget of almost nothing. Building one takes a semiconductor-scale factory.
The big, loud machine was the simple one, and it did its own explaining. The silent pane is the hard one, and it explains nothing at all.
Run that forward through a career and it repeats at every level. In 1999 you could follow the whole path with your eyes: the machine, the rack, the cable, the router at the end of it. By 2006 the machine was a line in a console and the rack belonged to somebody else. Now a model answers in a sentence and there is nothing to look at.
Every step made the thing easier to use and harder to see. The difficulty went up at each one. The chip factory, the datacenter, the training run: all of it heavier than what it replaced, all of it somewhere else.
The CRT was worse in almost every way that mattered and nobody should want one back.
Simplification is still the wrong word for what happened. The
complexity was relocated, into a chip factory, a datacenter nobody
visits, a training run nobody watches. It is all still there, and it is
bigger than it was.
You could understand an electron gun.
Follow the machine back far enough and you end up in foreign policy.
The finest layers are printed with light at 13.5 nanometres, thirty times shorter than the shortest colour the eye can see. It is made by firing a laser at tin droplets fifty thousand times a second. No glass bends it, so the machines work entirely on mirrors. One company sells them, ASML. One supplier on earth makes the mirrors, Zeiss, and ASML's own annual report says that without it the company would cease to be able to operate. The newest machines cost about four hundred million dollars each. Fewer than a dozen exist. They ship in pieces aboard several aircraft. The factories that use them are overwhelmingly in one country, Taiwan, and mostly at one firm, TSMC, which turns out something like nine in ten of the most advanced chips on earth.
Washington and The Hague now write rules about which of those machines may be sold to
Beijing.
The arrangement has a name, coined outside Taiwan: the silicon shield.
An industry the world cannot replace is a national security problem in a dozen
capitals at once.
The bet is that nobody wants it broken.
It was the television-shaped thing on the desk.
Now it is a matter of state.