“The future is already here, it’s just not evenly distributed.” People like me say it in meetings to sound well read. I used to credit Asimov. It’s William Gibson. Which tells you how much reading went into it.
It sounds profound because it feels true. It usually gets wheeled out for flying cars and robot butlers. I think it works just as well for downturns. The end of a boom arrives unevenly too.
China first
Chinese AI stocks have already crashed. Reuters reports that the STAR50, the index of China’s AI and chip leaders, fell 40 per cent in less than four months, wiping out about $480 billion. Unitree, the robot maker retail investors oversubscribed 8,000 times, has halved since August. A Shanghai fund manager expects the chipmakers to end up in a price war. “All bubbles inevitably burst,” he said.
Maybe it’s a blip. Some banks are already calling it a buying opportunity, and the Nasdaq has only dipped. But let’s say it’s the start. What happens next?
The after-party
When a hype cycle goes bust, the technology doesn’t disappear, lex Kevin Kelly’s No Extinction observation. The data centres are still there, the models still run, and the people who know how to use them are suddenly cheap and bored. What disappears is the business case.
That’s when it gets interesting. With no profit motive, people shoehorn the tech into places nobody would fund. Most of it is pointless. Some of it turns into the next thing. The fibre laid during the dot-com boom was dirt cheap after the crash, and Web 2.0 was built on top of it.
Traces of shoehorning
I think the shoehorning has already started. Two recent examples.
nesformer runs a Nintendo Entertainment System inside a stock Qwen language model. The console’s processor, memory and game cartridge are compiled straight into the weights. Nothing is trained. The model then “predicts” the game one token at a time, frame-perfect, about 150 to 290 times slower than a console from 1985. Speed is not the point.
bootgpt goes the other way. It’s an 11.5 KB neural network that boots on a 386 without an operating system and has exactly one job: to hallucinate the startup log of the machine it’s running on. Do androids dream of electric sheep? No. They dream of boot logs.
Neither will make anyone rich. Both are what people build when the question stops being “what’s the market?” and becomes “what happens if?”
Out of the ashes
None of this means AI is becoming a fad. The models aren’t going back in the box, and nobody is tearing down the data centres. What goes is the story that every chatbot is worth a trillion. What’s left is cheap compute, bored engineers and a lot of odd experiments, which is usually where the useful stuff comes from.
All of this has happened before, and it will happen again, ain’t that so admiral Adama? The leaves fall every autumn. Nobody calls the tree a bubble.