What's this about?
A financial rather than technical argument about the AI boom, which Gary Marcus amplified. The passage doing the work: OpenAI's equity, valued north of $850 billion, is "functionally the junior tranche of a capital structure whose senior claims — the take-or-pay compute obligations — exceed any revenue path management itself has articulated."
What is a "junior tranche" and what is "take-or-pay"?
When a company is funded by several kinds of claim, they get paid in order. Senior claims get paid first; the junior tranche — usually equity — gets whatever is left, which in a bad year is nothing. "Take-or-pay" is a contract where you commit to pay for capacity whether or not you use it, common in compute deals. Put together, the argument is: the compute bills are legally senior to the shareholders, they are larger than the stated revenue path, so in a shortfall the equity absorbs the entire gap.
What's the community saying?
- The most useful comment carefully separates two arguments people keep conflating: Marcus believes the technology doesn't work; this article mostly argues that even if it works and creates enormous value, the *timing* may not fit the capital structure.
- A third possibility gets proposed and upvoted: the tech works but not well enough to resist commoditisation — if open or on-device models satisfy 80% of consumers, the premium market shrinks to specialists.
- The counter to the doom reading: labs will be fine as long as investors believe, and they overwhelmingly do; drawing on traditional market wisdom fails because none of this is precedented.
- A sharper structural objection: the claim that large investors won't let their investment be seized amounts to small investors sheltering behind sunk costs, and those investors got large in a near-zero interest rate era that no longer exists.
- The inevitable "so should I move to bonds" sub-thread, and the reply that in a crash everyone wants cash, so bonds fall too — go look at 2008.