Imagine a scenario where you'd love to have access to a large number of my digital widgets, but they're expensive to make or buy, and a large number of them is really expensive. So you train an ML model on my things you can't afford to buy. It's still expensive, but that's a one time cost. Spend $5M training GPT-3, it's fine. Now you can sample from the space of my digital widgets. You have gotten a large number of widgets, just by throwing money at AWS. With money, you have converted my widgets into your widgets, and I'll never see a cent of it.
That's the issue. Content is expensive and it's still needed. Traditionally, I make content and if you want to benefit from my labor, you pay me. In the future, if you want to benefit from my labor, you pay AWS instead.
tl;dr The most significant equation for generative models is "$$$ + my stuff = your stuff"