An AI bubble requires actual AI capabilities to be built and to fail to be called that. A stock trading scam on AI stocks is just a stock trading scam.

36GW is 36 times current US corporate owned AI datacenter capacity. All of it made at TSMC which runs full tilt as it is. TSMC’s current plans is to 2x (over 3nm today) 2nm node by 2028. 4x by 2031 would still mean 2038ish to satisfy just OpenAI demand.

A big problem with having so many partners that OpenAI has no actual money to pay, is that no one has the confidence to place big TSMC capacity unless they are sure OpenAI actually buy from them, with money they get on time to buy from them. There are definitely consumer AI/gaming capabilities that can have more certain demand than the datacenter impossibilities.

Each additional announcement makes the joke more impossible to believe. CISCO did not do this during 1999 bubble. People just assumed its growth would continue.

Fine, OpenAI revenue growth may reach 3x at end of this year. At losses equal to its revenue number in 2024. OpenAI already has the most expensive models.