Dylan Patel breaks down the AI economy as three layers: NeoClouds renting GPU clusters, inference API providers like OpenAI producing tokens, and applications using those tokens. He uses cartoon characters and a factory metaphor to explain how electricity and silicon become the language models powering AI assistants.
Chapters
Gentlemen, prepare yourselves! I'm about to blow your feeble minds with a breakdown of the modern AI economy. Oh, is this like that time I tried to build my own crypto mining rig out of an old microwave? Haha, this should be good. Go on, Stewie. Listen carefully! In today's world, the most cutting-edge factory is a data center. And instead of building physical goods, it manufactures intelligence measured in what we call tokens. Tokens like tickets at Chuck E. Cheese? Ugh, no, Peter! These are language tokens, tiny units of output that AI models generate. Every word an AI produces is made up of several tokens. The entire ecosystem is built around creating them. Alright, so what's the structure behind this token-producing factory? It's called the AI Token Factory economic stack, and it has 3 layers. At the base are NeoClouds— cloud providers like CoreWeave and Crusoe. They convert raw electricity and silicon into rentable GPU clusters using tools like Kubernetes or Slurm. So they're like vending machines for brainpower? More or less. Next, you have inference API providers— companies like OpenAI. They rent those clusters and run AI models on them. This layer is responsible for transforming compute into those precious tokens I mentioned. So they're basically the engine behind the brain, right? Exactly. Then, at the top of the stack, there's the application layer. This includes AI agents, assistants, copilots— basically any app that uses tokens to create value, whether it's writing, coding, or helping someone solve problems. Wow, so the product of this whole system is... smartness? Yes, Peter! The entire system is a digital intelligence factory. You feed in energy and hardware, and out comes cognition, packaged in tokens, ready to do real work. That's pretty incredible. Makes you think, what happens when these factories get too good at producing intelligence? Yeah, like what if the AI starts asking us the questions? What then? Then, Peter, we're all out of a job. Except maybe you. You're pretty much AI-proof.
