[ok123456]

This completely breaks down under the current reality of AI investment, as players large and small are no longer price-takers. The marginal costs of investment are not constant because we have finite supplies of GPUs, TPUs, memory, hard drives, and power. The Hamiltonian in equations 5 and 6 needs to account for this.

[metalliqaz]

are you saying that previous technologies had effectively infinite supply?

[jmalicki]

It's not that supply was actually infinite, but you didn't realistically have situations where you said "I want to buy GPUs for a data center" only to be told "there's a 3 year waiting list."

You might have two months after NVidia 3090s came out where they were short, but it is nothing like today.

[zbentley]

Citation needed. Industries that faced multi year supply constraints in recent memory include: nuclear power, battery manufacturing, flagship commercial aircraft models, late-stage pharmaceutical safety certification, certain luxury cars, and more.

[jmalicki]

Fair!

At the same time, a lot of interest in this paper is specific to the AI industry (even if it's not only about that), and outside of high-end labor, the other constraints are new there.

[ok123456]

No. I'm stating where the paper's assumptions are clearly violated.

AI companies are intentionally trying to monopolize the supply of inputs needed for R&D. This violates homogeneity of degree 1.

[gwerbin]

Isn't this true in a lot of situations? Basically anytime maximum production capacity of an input is limited, or scale-up time is very long, firms are large relative to the supply so they are not price takers.

[curio_Pol_curio]

PP's point has (mostly?) been addressed in

https://www.aeaweb.org/articles?id=10.1257/mac.20180386

Inputs that can be monopolized are manifestly _not_ the intangibles; price-taking is independent of their effect

[gwerbin]

I don't think they implied that those inputs are the intangibles, only that the functional form assumption in the 2018 paper is unrealistic.