[JumpCrisscross]

> this situation is unlikely to get better in the short term because building cheaper per-token capacity is very expensive and time-consuming

They don’t need to fix it in the short term.

Look, this could be total nonsense. But what won’t happen is Anthropic or SpaceX disappearing inside a year. That was true in the 90s because the only cash flow going into those companies came from investors.

[marcus_holmes]

I notice you left out OpenAI from that ;)

Agree, some of these are valid businesses. But they are also massively overvalued on that underlying valid business, because of investor enthusiasm. When the bubble pops they are going to have real problems because of that overvaluation. Hopefully they survive, as a lot of the dotcom businesses did.

I think the real bloodbath will be the second-tier businesses that are mostly reselling cheap tokens to a market niche with custom prompts, and also massively overvalued as "AI businesses". And that kinda mirrors what happened in the dotcom bust - all the overvalued "webscale" businesses that hadn't really worked out a solid model yet went to the wall immediately

[JumpCrisscross]

> notice you left out OpenAI from that

OpenAI seems to have made debt-like commitments to spending on infrastructure. If those are indeed binding, they may have less flexibility than the others. (If Anthropic’s revenue growth stalls and its valuation halves, it should still be a going concern.)

[marcus_holmes]

Thanks for satisfying my curiosity anyway :)