The AI bubble bursts on layer economics
The AI bubble won't burst on valuations or spend. It will burst on whether the layer on top can pay for the layer underneath.
Healthy infrastructure takes a thin slice of the value built above it. A mature software company pays AWS about ten cents on every dollar it earns, and keeps the rest. The infrastructure gets rich and the business on top gets rich. That balance is what makes it last.
AI doesn't look like that yet.
The model layer sells close to $100 billion a year. For that to be the thin slice it is in cloud, the software built on top would need to be earning near a trillion. Set the bar at half that and it still isn't close. The application layer is earning low tens of billions.
So the slice is bigger than the pie. The model layer collects more than the applications above it can turn into revenue, and it is scaling faster than they can catch up.
The revenue that does exist is thin. Some AI products run a model bill as large as their revenue.
Underneath all of it sits hundreds of billions in capital spending, placed on the bet that the profit shows up later.
The bubble ends the day the software on top can pay for the model underneath and still keep a margin. Not before.
I'm watching that gap more closely than any benchmark.