I'm not so confident in that yet. If you look at the inference prices Anthropic charges (on the API) it's not a race to the bottom - they are asking for what I feel is a lot of money - yet people keep paying that.

Yeah, a collapse should only mean that training larger models become non viable right? Selling inference alone should still deliver profit.

Their employers are paying that money. The jury is still out on how wisely that money is being spent.