How is that compared to the speed at which money is thrown in them? And how much of the profit downstream they might indeed potentially capture with high retention rate?
Bubble can’t be assessed on mere quarter profit expansion over short period (relative to large company group life expectancy).
Don't you think that it's less than reasonable to suggest that the technology will take on the order of 20 years to mature into widespread adoption and profitability?
Then you should invest heavily in their IPO, I suppose.
You'll do much better than those of us who believe there is no way their financials can ever work.
I came to my conclusion a different way, I ran different models locally and measured the amout of resources required for basic things. From that I extrapolated some napkin unit economics, and I couldn't find any way to build a profitable business from it.
I don't assume I'm the smartest guy in the room though, so for all of our sake, I hope you're right.
I'm not sure if that is relevant here, but I've often seen people dramatically overestimate inference costs.
If we look at eg. https://inferencex.semianalysis.com/run/kimi-k31-pro-on-b300, it suggests that SOTA models are likely served in the API with profit margins in the ballpark of 98%.
In which case, the people being profitable won't be the generic big LLM companies. It'll be the people who use the LLMs, or (maybe) specialized AI companies that still have a niche and moat.