I think you are in for a rude shock - your expectation that there will come a reckoning where people are forced to content real prices of AI.. will never happen. It won’t.
In fact, there's a price war where some of the US inference providers are undercutting the pricing of Z.AI's own GLM hosting. Novita & AtlasCloud are both offering 8% and 5% discounts on GLM 5.2 respectively. GMICloud is charging 30% less - but getting so hammered with demand that it only has 80% uptime & 7 tokens per second, so you get what you pay for.
You can find a list of providers & their pricing through OpenRouter here:
Unless somebody will find a way to actually do training by just updating weights with new data you have to train from scratch and this is very costly process.
Saying that some lab is profitable because they are profitable doing inference while disregarding training cost is not fair assessment.
A product or a platform could give them a moat, but they don't quite have that yet although I think they are looking to get there