Actually there are two different things going on here
1. model layer companies: OpenAI, Anthropic
2. hyperscalers, data centres, hardware layer
When people say the bubble will pop, they usually mean the second one.
> I also think the narrative that these companies are profitable on compute is erroneous.
I don't think you read the Anthropic filing. It is substantially profitable if you exclude the one type of debt which was actually equity. It so happened that Amazon lent some money to Anthropic but it was tied to the valuation of Anthropic, which became a "problem" because Anthropic's valuation grew 10x. You wouldn't call this a loss in any other circumstance.
Anthropic filling shows negative EBITDA, how is that profitable? Revenue < (computing + personal + selling expenses). That is not a business, that is a money burning furnace.
That before even taking into account the huge amount of money they promised to datacenters (i.e. "build that for us, and we would rent it for 5 years", non cancellable).
But this is a bet we can settle right now: https://www.sec.gov/Archives/edgar/data/2130357/000162828026... , the ammended S-1 with Q3 data (that for some reason they omited from the S-1 that made big news), page 23:
(units in Kilo-millions)
Revenue in first 2026 9-months 33.2
Compute and infra expenses 18.8
Personel 4.8
Sales 5.5
Administrative expenses 5.3
Operating loss 1.2 (they are still losing money before ITDA)
But lets compare this with S-1 non ammended (https://www.sec.gov/Archives/edgar/data/2130357/000162828026...) Revenue in first 2026 6-months 16.3
Compute and infra expenses 8.9
Personel 3.1
Sales 3.2
Administrative expenses 1.7
Operating loss 0.7
Hmm, seems that doubling their revenue also doubles their expenses, causing double the operating loss.First 9 months minus first half:
Revenue in Q3 16.9
Compute and infra expenses 9.9
Personel 1.7
Sales 2.3
Administrative expenses 3.6
Operating loss ~0.6 (Remember, this is EBITDA)
You are lucky you didn't bet actual money.