John D Rockefeller stumbled on a winning strategy to an expensive problem. People were dropping wells everywhere they could see. They were looking for black gold. Most wasted money pursuing this. Some went belly up trying to get it.
But not old John. He figured out that you wanted to own the endpoints of oil production.
Wildcatting, as this oil welling drilling business was called, was risky. Sometimes you struck it rich, other times you didn't. But everybody had to get their oil refined. That's where the money was. And so, Rockefeller built a fortune.
This seems to rhyme with the new era of "AI"/ML models. New services like ChatGPT are blazing new paths.
But are any of these even profitable? ChatGPT says that each query costs them something like 5¢. That's much higher than Google's cost-per-query. You would need to sell at a higher price to justify that.
It feels like we might be entering a new Oil era. Much like oil, the lowest cost, best performing model will be picked at any time. The differences between models is almost interchangeable. They can be swapped almost at will.
In the past, API integration was difficult. But will AI chat models be as difficult to integrate? I have my doubts. Chat models have more lubricant, their UI is meant to replace replace hard word with softer, vague language. The friction to replace APIs just isn't there.
In the end, owning the end point may (still) be the most valuable piece.