In discussions about how token efficiency maximization could benefit Hyperscalers, I came across a particularly good analogy:

Imagine AI as a toll road. There are two types of companies on it:
1️⃣ Car manufacturers
They produce “cars,” i.e., the models themselves, responsible for getting you to your destination—cars are profitable.
2️⃣ Toll booths
They don’t care what car you drive. Ferrari, Toyota, or a 12-year-old used Honda—as long as you cross the bridge, you pay.
Now a change has occurred in the market:
Everyone suddenly realizes they don’t actually need a Ferrari for their daily commute, so they all switch to cheaper Hondas.
The result is:
Car manufacturers’ per-vehicle revenue collapses, but because cars are cheaper, people drive more (due to lower costs), and all those extra trips still have to pass through the toll booth.
👉 The toll booth’s revenue actually rises.
In the AI industry:
AI labs = car manufacturers
Hyperscalers = toll booths
In the coming months, we will enter a phase:
While the entire economy downgrades from a “Ferrari model” to a “Honda model,” it simultaneously increases token usage by 10 times.