They talk a bit about where the AI companies are supposed to get their revenue from. I’d like to see these analyses go a step further and talk about how those AI customers are supposed to come up with that money as well.
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?
Keep in mind wealth is not a fixed thing. If AI makes their businesses more productive, it would allow the creation of more wealth, giving them more money to pay for AI with.
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.
Indeed, it is hard to say exactly how much wealth the internet creates, but it's a lot. E-commerce alone is $30 trillion industry; some estimates put the total economic value of the internet at $200 trillion or higher.
That said, this came too late for investors in the dot-com boom. AI investors may find themselves in a similar situation.
It isn't exactly not a zero sum game either though. If you want to sell something you have to have willing buyers with either funds on hand or sufficient credit to make the purchase. So yeah that 6T gotta come from somewhere.
Customers coming up with it is basically fine*, but it's also why the SpaceX IPO chose to use "TAM" as a big number to dangle before investors: if you can do the work of {$6T + markup}/year worth of desk jobs, that's your customers.
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
The historical precedent is those people get other jobs, and/or everyone else in the economy gets richer faster than they get poorer.
If this does or doesn't happen with AI kinda depends how fast it makes multiple different sections of the population unemployed, rather than just one group like "software developer" or "vehicle driver".
you might make a hypothesis they'd offset operations costs like headcount for this but I think the counterpoint is stronger, if they benefit from increased productivity in your example, why not run more net production from same input.
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.
Ultimately I think they’re looking at the taxpayers to bail them out of this bubble. Everyone knows they over invested, no one wants to admit it because they’ll lose money.
What does this mean? I've been using AI as a replacement for both of those for longer than a year. If I need a PowerPoint file I have it generated off my tooling. Instead of fiddling to align text you say update slide 6 to look like this, strike this sentence. And you can add even more dynamic content. Like live application dashboards connected to real sources.
You can personally do that but I don't see how having ChatGPT means mass uninstalls of excel occur in enterprise. AI using PowerPoint doesn't mean powerpoint got replaced right.
it seems fairly obvious to me that once ai becomes as good as a worker, people will pay a workers wages to use it - 24/7, can't quit, can't complain that badly, no lunch breaks
I truly don't think anyone serious believes that, but if they pretend to believe it and preach it aggressively they might make out with an historic payday and stick the rest of us with the bill. People did the same with Crypto, with NFT's, with a bunch of things for as long as the ball is rolling.
Even if they were right... they don't really understand mass unemployment of that scale, right? Hungry people with a lot of time on their hands are known to get particularly inventive to entertain those who own too much.
AI is the second Industrial Revolution! You cannot use outdated notions. The nature of work and value creation itself is changing. Those who do not embrace AI will be left behind. Soon AI will accelerate in its ability to self-improve resulting in general artificial intelligence. This will result in not just our economy reaching the next level but will illuminate the very nature of knowledge and existence. We will ascend becoming beings made of pure energy.
To put this into perspective, healthcare, real estate and banking each earn ~$1.5T in revenue annually. This is REVENUE, not profit which is obviously lower than the revenue in every industry. I fail to understand how AI can achieve $6T in annual revenue when the TAM for healthcare and banking is a quarter of that in the US.
Then again, investments in AI are 1-2 % of worldwide GDP I think, if you think that this technology can probably increase efficiency at least a couple of times in anything that has to do with information processing it doesn't seem so ridiculous. The electric grid, railroads and other large-scale infrastructure required similar amount of investments, though they took much longer. I guess the main risk here is the speed at which everyone races to capture this market as unlike electricity or railroads this revolution is entirely digital so competition happens nearly on a global scale and everything is much faster. That is probably the main risk, going this fast will definitely cause overshoots in one or the other direction.
> 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
This is a 2031 projection of a 6T market, but somehow the article editorializes this into 'justifying' a boom?
I'm fairly confident that the crash will happen because token prices fall through the floor due to competition from the open source models. This will destroy the valuation of the AI firms and their ability to raise capital, so they'll stop spending hundreds of billions per annum on training.
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.
Such an interesting dilemma. Arguably an incredibly transformative new industry for which the market cannot fundamentally bear the costs. (or IDK, does it actually work now?) How will it play out??
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?
Zitron has been talking about this for some months now. The numbers fluctuate, I think $6T is the highest I've seen but low-end is about $1T that doesn't exist today.
I think Zitron's biggest mistake is making too many concrete and confident assertions without enough backup. Yet the AI safety doomers have basically never been right either, but nobody seems to mind that; clearly they'll be right when they're right, just wait for it?
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.
> Yet the AI safety doomers have basically never been right either, but nobody seems to mind that;
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 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.
I wish people would realize that Zitron is nothing but a grifter who saw an avenue for content and went all in.
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
The high end sectors using these servers are paying way more than joe consumer and his 30 bux a month. We are already seeing justification for the boom, and I'd expect even more advancements to come in multiple sectors. That number is missing from the article.
That segment would include innovations in search, advertising, autonomy and physical AI, analysts at Boston-based Bain said.
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.
it does seem a lot like the whole AI thing is the market staying irrational a lot longer than the naysayers can afford not to go along with it, but still, nobody can stay irrational forever.
They talk a bit about where the AI companies are supposed to get their revenue from. I’d like to see these analyses go a step further and talk about how those AI customers are supposed to come up with that money as well.
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?
Keep in mind wealth is not a fixed thing. If AI makes their businesses more productive, it would allow the creation of more wealth, giving them more money to pay for AI with.
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.
this also happened with the internet. 6T is possible only if AI creates more wealth. same as with internet. Its not a zero sum game.
Indeed, it is hard to say exactly how much wealth the internet creates, but it's a lot. E-commerce alone is $30 trillion industry; some estimates put the total economic value of the internet at $200 trillion or higher.
That said, this came too late for investors in the dot-com boom. AI investors may find themselves in a similar situation.
It isn't exactly not a zero sum game either though. If you want to sell something you have to have willing buyers with either funds on hand or sufficient credit to make the purchase. So yeah that 6T gotta come from somewhere.
Customers coming up with it is basically fine*, but it's also why the SpaceX IPO chose to use "TAM" as a big number to dangle before investors: if you can do the work of {$6T + markup}/year worth of desk jobs, that's your customers.
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
> Do we really expect all the world’s corporations to suddenly up their annual spend by 1,000% on average? Why?
If they keep replacing employees with AI, that is indeed the direction.
But if all those peolple are unemployed will they still be able to buy all the products the corporations sell…..
The historical precedent is those people get other jobs, and/or everyone else in the economy gets richer faster than they get poorer.
If this does or doesn't happen with AI kinda depends how fast it makes multiple different sections of the population unemployed, rather than just one group like "software developer" or "vehicle driver".
It’s true - but then where do folks get money to buy from the businesses?
Eventually this merry-go-round runs into some issues if we don’t figure out a way to make this huge productivity boom work.
you might make a hypothesis they'd offset operations costs like headcount for this but I think the counterpoint is stronger, if they benefit from increased productivity in your example, why not run more net production from same input.
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.
In a recession, too. You have to ask where the money even enters the funnel.
Ultimately I think they’re looking at the taxpayers to bail them out of this bubble. Everyone knows they over invested, no one wants to admit it because they’ll lose money.
> enterprise productivity tooling market
AI is not a replacement for powerpoint and excel. You can't substitute those.
White collar worker salaries is ~$30T+ global.
What does this mean? I've been using AI as a replacement for both of those for longer than a year. If I need a PowerPoint file I have it generated off my tooling. Instead of fiddling to align text you say update slide 6 to look like this, strike this sentence. And you can add even more dynamic content. Like live application dashboards connected to real sources.
You can personally do that but I don't see how having ChatGPT means mass uninstalls of excel occur in enterprise. AI using PowerPoint doesn't mean powerpoint got replaced right.
Please don't cover up one insane claim with an even less plausible one.
I'm very confused by what you mean.
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?
it seems fairly obvious to me that once ai becomes as good as a worker, people will pay a workers wages to use it - 24/7, can't quit, can't complain that badly, no lunch breaks
productivity tooling replaces employees
> people will pay a workers wages
Who re these people?
Total software developer comp (worldwide) is ~$1T
anecdotally My AI aka personal productivity spending has gone up something like that
I truly don't think anyone serious believes that, but if they pretend to believe it and preach it aggressively they might make out with an historic payday and stick the rest of us with the bill. People did the same with Crypto, with NFT's, with a bunch of things for as long as the ball is rolling.
Because they’ll be able to just fire all their employees and redirect the spend to AI. That’s the current delusion.
Even if they were right... they don't really understand mass unemployment of that scale, right? Hungry people with a lot of time on their hands are known to get particularly inventive to entertain those who own too much.
Seriously. We’re supposed to believe this AI boomer rhetoric that it won’t take jobs. In order for companies to pay for it, it must take out jobs.
AI is the second Industrial Revolution! You cannot use outdated notions. The nature of work and value creation itself is changing. Those who do not embrace AI will be left behind. Soon AI will accelerate in its ability to self-improve resulting in general artificial intelligence. This will result in not just our economy reaching the next level but will illuminate the very nature of knowledge and existence. We will ascend becoming beings made of pure energy.
This really is the only scenario that justifies current valuations.
To put this into perspective, healthcare, real estate and banking each earn ~$1.5T in revenue annually. This is REVENUE, not profit which is obviously lower than the revenue in every industry. I fail to understand how AI can achieve $6T in annual revenue when the TAM for healthcare and banking is a quarter of that in the US.
Then again, investments in AI are 1-2 % of worldwide GDP I think, if you think that this technology can probably increase efficiency at least a couple of times in anything that has to do with information processing it doesn't seem so ridiculous. The electric grid, railroads and other large-scale infrastructure required similar amount of investments, though they took much longer. I guess the main risk here is the speed at which everyone races to capture this market as unlike electricity or railroads this revolution is entirely digital so competition happens nearly on a global scale and everything is much faster. That is probably the main risk, going this fast will definitely cause overshoots in one or the other direction.
Enjoy the bredth of choice you have with LLMs. Most models will fail and disappear, and those that don't will become a lot more expensive.
If the market crashes, and I can buy a TB of TPU on the cheap, I can host 4.1 flash myself. :D
I'm fairly confident that the crash will happen because token prices fall through the floor due to competition from the open source models. This will destroy the valuation of the AI firms and their ability to raise capital, so they'll stop spending hundreds of billions per annum on training.
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.
Such an interesting dilemma. Arguably an incredibly transformative new industry for which the market cannot fundamentally bear the costs. (or IDK, does it actually work now?) How will it play out??
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?
Zitron has been talking about this for some months now. The numbers fluctuate, I think $6T is the highest I've seen but low-end is about $1T that doesn't exist today.
His latest newsletter (from today) also talks about this: https://www.wheresyoured.at/dead-money/
Knowing that Zitron wrote about this makes me think that 6T will actually be reached, given Zitron's abysmal track record.
I think Zitron's biggest mistake is making too many concrete and confident assertions without enough backup. Yet the AI safety doomers have basically never been right either, but nobody seems to mind that; clearly they'll be right when they're right, just wait for it?
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.
> Yet the AI safety doomers have basically never been right either, but nobody seems to mind that;
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.
Zitron is the Greta of tech.
What's that mean?
Obviously means greta and this guy are negative indicators. They are so wrong you can do the opposite of what they tell you and profit reliably.
> Zitron's abysmal track record
https://danluu.com/zitron/
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.
I wish people would realize that Zitron is nothing but a grifter who saw an avenue for content and went all in.
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
It's gonna have to be so good that it creates entire industries that didn't exist before. That would be awesome, but I'm not seeing it so far.
The high end sectors using these servers are paying way more than joe consumer and his 30 bux a month. We are already seeing justification for the boom, and I'd expect even more advancements to come in multiple sectors. That number is missing from the article.
That segment would include innovations in search, advertising, autonomy and physical AI, analysts at Boston-based Bain said.
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.
The sooner it pops the better
it does seem a lot like the whole AI thing is the market staying irrational a lot longer than the naysayers can afford not to go along with it, but still, nobody can stay irrational forever.
as ai becomes an actual replacement for workers, costs will rise to what you would pay a worker, and $6T will be easy
Feels right and sustainable, ballpark.
Them AI must make more jobs than the opposite, nobody unemployed will spend cash with shit
not really, stop these BS