[conductr]

Growth of what exactly? AI doesn't have the normal leverage factor that software usually does where a simple codebase can drive a billion dollars of subscription revenue with 90%+ gross margin. There's no eventual state where the capex is in place and the margins flip. They're in the datacenter business, which is real estate, with tenant improvements consisting of rapidly depreciating/obsoleting equipment. These margins have no path to flip around and allow for a huge amount of revenues to flow through. If they start testing price sensitivity in the way that would justify the valuations, it will just accelerate the transition of AI from datacenter to local.

>Experts continue to warn of an AI bubble, noting the enormous and widening gulf between company valuations and their comparatively measly profits

[DangitBobby]

I think your analysis assumes current closed SOTA models have no moat, which isn't a given considered the protectionism the Trump administration is already considering for US AI firms.

[waterTanuki]

It is a given. Unless the Trump admin figures out how to ban 1s and 0s, you can't stop Chinese labs from open sourcing their models. They can certainly ban new providers from using the Chinese models, but they can't ban Canadian ones, European ones, other Asian countries etc. from setting up provider proxies and bypassing those rules. Then what, great firewall of USA? Block off all internet traffic to other countries? Bye bye economy.

I think the burden of proof is on the closed SOTA model providers to prove they have a moat, because common sense indicates they don't.

[DangitBobby]

Yes, banning 1s and 0s is completely impossible and unprecedented /s. I wouldn't choose either as a default until the dust settles. Either way, I wanted to point out the unstated assumption. Whether you personally believe it's a foregone conclusion or not, it's an essential part of the logic chain that not everyone will agree to.