The point is that the providers aren't burning money by subsidising inference costs. On the contrary, if this article is to believed they're charging healthy margins on it.
So there are two answers: for the model providers, it's because they're spending it all on training the next model. For the API users, it's because they're spending it all on expensive API usage.
The point is that the margins aren't "healthy" for the industry if their users can't be profitable, because if that's the case, the users will all go out of business, and the providers will stop being able to charge anybody anything, margin or no.