The AI that we’ll have after AI

Cory Doctorow argues that when the AI bubble pops, the useful residue will be cheap GPUs sold at firesale prices, skilled applied statisticians looking for work, and open source models that can be optimized to run on modest hardware. He contends that AI companies spend heavily to signal investment rather than to build products, citing a $45b/year industry-wide revenue figure against spending of most of a trillion dollars in capital expenditures. He points to DeepSeek, whose launch knocked $589b off Nvidia's stock price, and to Pete Warden's demo of a voice assistant running on a chip costing "low single digit dollars."