Comment by baggachipz
11 hours ago
Hate on him all you want (he's gotten very repetitive for the sake of subscribers and reads), but the basic premise that the modern AI industry is a circular-dealing, point-of-diminishing-returns grift still holds. The bubble is here and the longer it inflates, the worse the pop will be. Sure the goalposts have moved, but the basic numbers don't math.
The current cross-deals aren't purely circular; quite a bit of revenue is, in fact, flowing into the AI industry from the outside (you know, via customers). What's more notable is the shared risk, as the deals tie multiple companies across the chain to a set of shared bets.
If the frontier labs suddenly found themselves unable to compete with cheap open weight models running on widely available compute, then one might expect the frontier labs to be the only ones exposed to that risk, while a chip-manufacturer like nvidia could thrive in either environment. A partnership or commitment from the chip manufacturers to the frontier labs could change that. Whether that kind of inescapable connection exists is hard to predict without a lot of specific modeling, and at this point I'm inclined to think that neither chip nor datacenter demand is going to drop any time soon, and that the commitments would be unwound before a company like nvidia is threatened.
> The bubble is here
Zitron implied 2 years ago that OpenAI would collapse by now. How's that bubble popping going? All NVDA+memory co+frontier lab numbers are accelerating
The thing about bubbles is that everyone who thinks they’re going to pop look crazy until they pop.
And to be clear, a bubble popping doesn’t mean that AI goes away forever.
What it does mean is that we’ll see some kind of economic crash or recession, and we’ll probably see at least one big company fail or go bankrupt/restructure.
OpenAI is the company in most obvious peril.
I happen to think that Nvidia is in a more perilous position than they appear. Their hardware advancement pace is relatively weak and they’re in a crypto-like hardware bubble where they’re one technology breakthrough away from a complete collapse in demand for their AI data center solutions. They’re also doing a lot of sketchy hardware financing schemes.
Comparing to other bubbles in the past, this one could still have something like 2 years left. Patrick Boyle has a video on the topic, in which he's also very careful to point out he could easily be wrong.
The market can stay irrational for quite a while.
By all means, keep betting on this being a wild success and we'll see who's right in a little while. If you don't think this is a bubble then you're in for a ride.
Look if we are going to raid pensions and 401Ks to prop up the valuation targets of Anthropic and OpenAI longer, then the bubble can stretch further and further, but eventually datacenters have to get built and powered. It's the power generation part that no one talks about. We simply don't have enough power in the United States to scale at the rate that Anthropic and OpenAI need to prop up their absurd valuations. AI is real, but these valuations are not.
TL;DR: Ed is directionally correct, but it's anyone's guess as to the exact timing.
In the meantime I'm not going to complain about subsidized credits from the big labs. :-)