Are OpenAI and Anthropic losing money on inference?

Doesn't he have an incentive to make it look like that, though? The way he phrased it, that they are losing money because people use it so much, makes it seem like Pro subscribers are some super power-users. As long as inference has a nonnegative, nonzero cost, then this case will lose money, so Sam isn't admitting that the business model is flawed or anything