Do they get more out of it than it costs, or are they still in the "people are just giving us money in the hopes that one day it turns a profit even though we're not charging nearly enough to make a profit" phase?

You're describing the AI companies and their business model.

I'm answering to that cost being a problem regarding "what prevents 100 Billion ChatGPTs from using any protocol?" - the context I have in mind for the above being scammers, political manipulators, spam, and people like that using ChatGPT/LLMs to take advantage of various protocols for profit (and the 100 billion figure being a figure of speech meaning "very many").