His latest newsletter (from today) also talks about this: https://www.wheresyoured.at/dead-money/
I kinda hope corpos with a bunch of money and plenty of normies adopt AI so developers don't get bled (as) dry, esp. given the recent OpenAI changes which I feel like Anthropic is soon to follow.
Edit: explanation of the downvotes would be nice. Dan Luu made a fairly factual article there that explores some of Zitron's claims. And personally I don't see how it's bad to wish for wider adoption of AI so the big orgs get their fill of profit (since enough inference could help fund R&D better) and developers don't have to bear that burden.
Just for fun, I'll join in myself and make two bold half-hearted predictions that will surely be wrong:
- Zitron is wrong about the details, but right about the ultimate fate of the AI companies. Turns out AI is useful after all, it just isn't able to justify trillions of dollars in expenses. Shit falls through hard. Probably not an AI winter, but more like the dotcom bubble, as many have already predicted for years. Everything looks rosey right up until the first domino falls.
- The AI safety doomers are right, just not any time soon. Turns out our depth of understanding what intelligence really means was severely lacking; it's hard to quantify and we never had so many things testing it. Eventually, we really do discover why LLMs emergent capabilities really are inherently limited in ways that just make their ability to meaningfully "escape" less scary, and maybe architectures emerge without these limitations, just not yet. The fight with LLMs and malicious use winds up looking a lot more like a supercharged version of the current fights we have with dumber bots. Everyone winds up having to fight LLMs with LLMs, and this new world eventually pushes both good and bad new legislation in its wake, finally actually pushing responsibility for AI misfires, but probably in a way that is at least as bad as it is good, creating chilling effects and harming competition unnecessarily, because that's just how things go. Maybe we eventually do crack true super intelligence where it's a full superset of human mental facilities, and then geopolitics maybe gets as interesting as it ever has been in history. Probably not in a way that is fun for people living through it.
I can't blame Zitron. It's fun to pretend to be able to predict the future. I still broadly agree that the financials don't really seem to make that much sense even if he clearly keeps getting the details wrong.
I don't think that nobody seems to mind that. Pro-AI people make fun of doomers because of their failed predictions of apocalypse at least since GPT-2 days.
I don’t even disagree with some or even a lot of his takes, it’s just he is singularly a content generator with a track to run on. Look at his previous work, it’s much of the same kind of attempts at being relevant. To me he’s nothing more than a drive time radio host, shouting about something just to fill the air
Consumer-focused services, which includes subscriptions and advertising revenue, is seen to contribute $200 billion to $400 billion
So not $6T of ARR subscription sales, but $6T of "economic enablement". By 2031, the global economy is estimated to add another $40T of economic activity to GDP. So AI needs to account for 15% of that.
For example, the article suggests $1T from enterprise productivity tooling. A quick Google suggests the global enterprise productivity tooling market is only worth about $0.1T right now.
Do we really expect all the world’s corporations to suddenly up their annual spend by 1,000% on average? Why?
OTOH, this price crash will increase demand for inference. Maybe some data centres will go bankrupt, but the scavengers won't be selling off used TPU's for cheap, they'll operate those data centers to sell you tokens for cheap.
> forecasts that annual spending on AI infrastructure might hit $1.5 trillion by 2031.
> ...
> sustaining this level of investment would require an AI market approaching $6 trillion annually
I don't care how good AI is at something like software development or whatever task. If customers were paying real prices for usage would any of us be using it?
Any data out there on how far off enterprise API pricing is from actual compute COGS for these models right now?
AI is not a replacement for powerpoint and excel. You can't substitute those.
White collar worker salaries is ~$30T+ global.
My point was that AI isn't some productivity software thing. The addressable market is white collar work that can be done on a computer.
You think its implausible for AI to replace some fraction of white collar workers? To me this is directionally exactly where this is heading?
Still though, the thesis seems based on an assumption that the companies would experience either major growth or major efficiency gain, neither of which really pass the smell test to me. Option 1 growth - money has to come from somewhere ultimately? Option 2 efficiency - put into profit rather than simple offset to AI provider?
Or the AI providers get companies hooked with subsidized pricing that unwinds later and you end up with the Uber/AirBNB situation of market capture and good product reverting to mean level of price/quality over time. Tinfoil hat thinks it ultimately gets squeezed out of retail investors in index funds.
This of course hinges on the big if: does AI create wealth? I think yes, but maybe not $6 trillion of it in five years.
That said, this came too late for investors in the dot-com boom. AI investors may find themselves in a similar situation.
However, customers coming up with the money is the easy half of the problem. The hard part is having them pay you.
Wikipedia's valuation is on the small side of {number of people on the internet worldwide} * {peak price of Encyclopaedia Britannica} † despite containing a lot more content.
LLMs can only chase high margins while they're ahead of their competitors; the moment the investment stalls, open weighs catch up, they loose their edge. Cloud compute providers still win on supplying inference, but there too competition will likely lower margins, not justify huge valuations.
NVIDIA may remain as valuable as today, but their P/E ratio will likely shrink a lot even while keeping up the price.
Model makers still have a problem here even if the tech is so useful it rapidly grows the economy: "free" is right behind them almost immediately after they stop.
* under the assumption the tech is real; there's plenty of people here on Hacker News who still dismiss it as a "scam"
† that would be about 13 trillion USD, from what I read
If they keep replacing employees with AI, that is indeed the direction.
Eventually this merry-go-round runs into some issues if we don’t figure out a way to make this huge productivity boom work.
productivity tooling replaces employees
Who re these people?
mkrishnan•47m ago