AI Demand Is Slowing Faster Than Wall Street Thinks | Jesse Felder

The guest argues AI demand forecasts built during a subsidized "token maxing" phase overstate how much data center compute the market actually needs.

Token-based pricing exposes AI's true cost, and the speaker says demand is already dropping off as businesses face real expenses.

On-device AI on machines like Apple's Mac Mini reduces dependence on hyperscale data centers, lowering projected infrastructure requirements.

A reported data center project hold at a customer's request is read as an early signal that hyperscalers may be overbuilding.

The speaker compares AI to the railroad boom, which turned to bust once builders recognized they had built more capacity than demand required.