Inference seems profitable and scalable enough, given time... The real cost seems to be in growth and training future models. Ergo, a slowdown can only improve the financial position of OpenAI and Anthropic (which don't seem stellar just now).