When the author leads with a part I know a few things about (the 2000s mortgage market) and gets it so wrong, that kind of makes it hard to put much credence in the point they actually want to make later.
Well, as an escape hatch you could say that the author merely says that the Big Short made these claims (which is true), and not that the claims themselves survive contact with reality. But that would be a lame cop out.
1. LLMs are remarkable and they've uncapped a supply/demand loop for software that has previously been much more tightly constrained than anyone realized. It turns out that if software is much cheaper and faster to make, people find ways to use a lot more software, so much so that the world is temporarily completely out of all the parts you need to make the machines that turn electricity into software.
2. The leading tech companies have spent wildly on something that seems likely to turn out to be a commodity that sells for a few points over what it costs to provide it on the expectation that the gains in software developer productivity would also apply to every other industry in a reasonable time, replacing human workers and increasing productivity.
Recklessly taking a trillion+ in debt to corner the market, only to find you can't actually corner it and can't built a real moat with this technology as long as everybody knows how it works (and everybody that wants to know, knows how it works), seems precarious, to me.
They have to make a lot more than $100/month off of everyone using their services for this to work out for them, and nobody wants to spend a lot more than $100/month for these services. People start looking around for alternatives the moment Anthropic says, "Well, first one's free, but we're going to take away the best model on the subscription plans pretty soon, of course."
Ed may be wrong on some points. But, it's hard for me to look at how much money the big guys have spent and not wonder, "Who's going to buy the services at the prices they need to charge?" It isn't going to be me.
He (and I) might have been wrong on how useful the LLMs could get, but time isn't proving him wrong on his claims that the usefulness doesn't justify the investments around LLMs
It is the fundamental question of is there enough willing buyers. And I am doubting that very much. Next step is to question if there is enough willing buyers and they are replacing white collar workers. Do these customers then have enough willing buyers? Is building data centres and infra to service them a viable way to replace most of the economy?
"It is difficult to get a man to understand something, when his salary depends on his not understanding it."
Hard to know of Zitron actually believes what he writes, is trapped in some type of mania, or keeps doing the same bit to pay the bills. Probably a mix of all 3.
He's a grifter who uses his followers to make money. He tells his followers stuff that he likely doesn't believe in himself but his followers want those things to be true.
To me the main question is what happens after that..
We all move to frozen open source models running on 2nd hand Oracle/Coreweave GPUs? 10 years before someone dares make another training run?
Culturally, do we all collectively sober up once money dries and hallucination are still here? Pendulum swing, AI consideredharmful moment? How to promote healthy use when cognitive surrender is so engrained in us?
What happens if there's a new GPT2 scale (i.e. not astroturf/mass histeria marketing) breakthrough?
I find that modern harnesses mostly get rid of hallucinations: ask your agent to provide quotes and citations and tests for everything, and have another agent cross check that.
Same way we used to do that as humans: I vaguely remember a case or an anecdote from history, but before I use it in a text, I go and look up whether my memory is playing tricks on me.
I mean the companies are delaying the inevitable. We have companies putting limit on the monthly token spend. Just 1% (you see what I did) of the companies can spend fearlessly on AI. In time most of the companies will be burnt out of their funding. I hope AI can make a case to be affordable.
Income by implemented features, losses by bugs and savings by personal. Sticks and stones may break even my bones, but ledger words will never hurt me.
Zitron is literally the worst person to raise alarms about the financials of the AI ecosystem because he's so hyberbolic and pollutes his own arguments with nonsense.
Take:
> When somebody decides to build an AI data center, they form a special purpose vehicle (much like a CDO), which then raises debt, in some cases slices it into tranches and, in most cases, sells them to institutional investors, asset managers or banks.
This is just such a weird and wrong comparison. A CDO's assets are other people's debt claims. The same mortgage bond could be split among many CDOs at once, those CDOs could be re-tranched into further CDOs, and thanks to credit default swaps, synthetic CDOs could reference bonds nobody in the deal actually owned. So basically exposure to a fixed pool of mortgages could be manufactured without limit.
A data center SPV's assets are the building, the power interconnect, the GPUs, and the customer contract. If the SPV fails, the loss is limited to what those things are actually worth. There are no multipliers as there are with CDOs.
Later in the post, Zitron even concedes this:
> What differs this from the subprime mortgage crisis is that the systemic risks aren’t driven by derivatives or complex financials but by the sheer scale of costs to build an AI data center, a catastrophic misunderstanding of the AI industry itself and the dangerous lending standards of private credit.
He claims this isn't important:
> When every single debt deal is over $500 million and usually numbering in the billions, we don’t need a vast web of different contracts to create a systemic risk, just clusters of projects that either fail to keep up with their SPVs’ debt or bonds that go unpaid by destitute or defunct data center developers.
But here's the thing: systemic risk isn't a function of how big the losses are. Instead, it's a function of who takes the losses and whether they propagate.
Equity holder losses just get absorbed by equity holders. What happened in 2008, on the other hand, was that the losses hit leveraged intermediaries funding long assets with overnight money, so one firm's distress became another firm's funding withdrawal.
Big deal sizes don't create that type of situation. A $10 billion SPV default is a $10 billion loss distributed across whoever bought the debt.
He brings up Lehman but that's literally the worst example for his argument. Lehman's losses were trivial against its $600 billion balance sheet. It failed because of a funding run. Repo counterparties refused to roll, the clearing banks demanded more collateral and prime brokerage clients pulled their balances. This doesn't happen in an SPV because SPV debt is term debt. It's sized and dated to match the asset. There are no runs on a term loan. When an SPV breaches its DSCR defaults, the lenders take the assets. It's not pretty, but it's contained. It can't spread beyond its own confines and multiply because there is no maturity mismatch, which is what killed Lehman.
What happens if the assets collected drop in value as they get repoed? Wouldn't the lender now also be in harms way and in turn have issues financing themselves?
> Wouldn't the lender now also be in harms way and in turn have issues financing themselves?
Yes, they could be. There are potential multipliers here, and one failed project could cause other projects to be marked down, even if they're not failing.
But the mechanisms by which $1 trillion in data center debt could become multi-trillion dollar write downs just doesn't exist the same way it did with subprime mortgages.
In the mortgage market, credit default swaps let the same collateral be referenced without limit, CDO-squared structures re-tranched the losses into concentrated wipeouts, and the paper was on the balance sheets of firms funding 30 year assets with overnight repo.
A derivative market for individual SPVs where there's a mismatch between the debt term and the asset just doesn't exist in the data center market.
Pensions funds that are involved have 30 year liabilities and no redemption pressure. Insurers don't have run risk. Banks hold the debt against capital buffers and they have access to the discount window. The riskiest debt holders are levered private credit and semi-liquid retail vehicles, but even in these cases, you're looking at leverage around 2:1, not 30:1.
None of this is to say that an AI bust couldn't cause widespread pain. Just look at how much of US equities are dominated by AI-linked companies. But the issue is that Zitron is just wrong about the mechanisms and magnitudes. He's trying to paint a 2008-like scenario because it's much scarier.
Ed Zitron is right about the need for a market correction, but he's fundamentally wrong on AI itself. He'll get his 15 minutes of "see I told you so" when the market corrects, but over the long term he'll be proven wrong.
Ed's primary gripe is that he thinks the business models aren't viable for profitability. But Google's AI infrastructure buildout is already spending less then the depreciation value of the hardware, meaning it's inevitably going to become profitable - at least for Google. Microsoft has since adopted the same approach that Google is using, focusing on faster and more efficient models in order to reduce costs.
Ed won't acknowledge that the paradigm shift for programming and SWE has already happened. He won't acknowledge that roughly 30% of radiology labs in the US and 40% of dental practices have adopted AI.
I personally think Ed is digging a hole he won't be easily able to climb out of.
> Can you make a falsifiable prediction without moving goal posts and ambiguous timelines?
I find this demand dishonest. The inherent nature of bubbles is that you cant predict when they pop. The more they grow, the more damage they cause, but it is impossible to guess what exactly triggers the pop.
So yes, ambiguous timeline is actually correct way to phrase things. Anything else is pure gambling.
I'm a SWE and use AI all day every day. I also acknowledged the very real rollout of AI in other facets of the economy besides SWE. But a market correction is inevitable.
We are in the wild west of AI. No one truly has agentic workflows "down pat" yet. We have no idea which AI workflows, pipelines, architectures, and platforms are going to stick around long term. For all we know a kid in high school can be the one who comes up with a SWE workflow that FAANGs and the rest of the tech sector ends up adopting.
Recent numbers from Google suggest that Ed Zitron might very well be wrong about the data center build-out, which is the single biggest crux of his argument. But even if he's wrong on that, he'll still get his 15 minutes of "see I told you so" (while still being wrong) because even if the data center rollout doesn't implode from assets depreciating too quickly, a correction will still need to happen on the dirt end of the shovels. Even if you don't have too many shovels (which Ed thinks we do), you still need to be digging in the correct patch of ground to strike a gold vein. There's going to be winners and losers in this thing, and the market correction will happen once the biggest investors and hedge funds have correctly identified who the losers are.
What I can't do is provide a timeline. I don't claim to be a prophet. However, if Ed is correct on the hardware buildout, the sheeit will hit the fan once amortization schedules hit critical mass. Otherwise it will be whenever the big money learns who the winners and losers are on the implementation rollout. That can happen at any time. It's perfectly normal and is not a doomsday clock. That's why it's called a market *correction*.
That is the quote of famous financial advisor made in 1986. It keeps being repeated, because bubbles and their pops and irrational markets are frequent enough.
There is nothing controversial about claim that markets are irrational, bubbles pop impossible to predict while bubble being visible. Previous bubbles were the same.
The problem with this type of thinking is that it deliberately ignores the times that bubbles were predicted incorrectly.
Imagine there was a guy. He kept predicting that a war may happen in 1 year. Keeps getting it wrong. In 59 a years a war does happen. He then retroactively justifies his wrong predictions by claiming “but people are irrational and it’s hard to predict”.
Past bubbles I read about were recognized correctly for years before they popped. The pop was hard to predict, the bubble was routinelly recognized. These were not wrong prections, these were correct assesments of the situation. Bubble exists before it pops.
Markets being irrational is again, nothing controversial.
That is history. Your war predicting guy is a made up analogy. And he may just be correct about his assesment of the situation - war is being actively planned and prepared, actors want it and we are all waiting till Trump or Hitler or whoever decides that "now".
Nobody with a minimum of credibility says that. Everyone says this is clearly a bubble and yet it could go on another couple of years.
Nobody is timing it. But if you want to deny the obvious by numbers go ahead
> Newsletter: Based on estimates of their burn rate and historic analyses, I hypothesize that OpenAI will collapse in the next 12-24 months unless it raises more funding than in the history of the valley and creates an entirely new form of AI.
He posted this in 2024 June btw.
He certainly doesn’t believe that OpenAI has a new form of AI. So where’s the collapse? Since then the revenue growth was exponential.
Zitron's posts reminds me of all the analysts who predicted the collapse of TSLA based on the fundamentals, and recommended shorting it. How does that saying go? "markets can remain irrational a lot longer than you can remain solvent".
There is an immense, coordinated, desperate effort to keep the biggest party in the history of the financial system going, to pour enough fuel into the engine that escape velocity may be achieved against the forces of mathematics.
When the author leads with a part I know a few things about (the 2000s mortgage market) and gets it so wrong, that kind of makes it hard to put much credence in the point they actually want to make later.
Well, as an escape hatch you could say that the author merely says that the Big Short made these claims (which is true), and not that the claims themselves survive contact with reality. But that would be a lame cop out.
I don't think Zitron will ever admit he's been wrong about LLMs
Two things can be true:
1. LLMs are remarkable and they've uncapped a supply/demand loop for software that has previously been much more tightly constrained than anyone realized. It turns out that if software is much cheaper and faster to make, people find ways to use a lot more software, so much so that the world is temporarily completely out of all the parts you need to make the machines that turn electricity into software.
2. The leading tech companies have spent wildly on something that seems likely to turn out to be a commodity that sells for a few points over what it costs to provide it on the expectation that the gains in software developer productivity would also apply to every other industry in a reasonable time, replacing human workers and increasing productivity.
Recklessly taking a trillion+ in debt to corner the market, only to find you can't actually corner it and can't built a real moat with this technology as long as everybody knows how it works (and everybody that wants to know, knows how it works), seems precarious, to me.
They have to make a lot more than $100/month off of everyone using their services for this to work out for them, and nobody wants to spend a lot more than $100/month for these services. People start looking around for alternatives the moment Anthropic says, "Well, first one's free, but we're going to take away the best model on the subscription plans pretty soon, of course."
Ed may be wrong on some points. But, it's hard for me to look at how much money the big guys have spent and not wonder, "Who's going to buy the services at the prices they need to charge?" It isn't going to be me.
He (and I) might have been wrong on how useful the LLMs could get, but time isn't proving him wrong on his claims that the usefulness doesn't justify the investments around LLMs
It is the fundamental question of is there enough willing buyers. And I am doubting that very much. Next step is to question if there is enough willing buyers and they are replacing white collar workers. Do these customers then have enough willing buyers? Is building data centres and infra to service them a viable way to replace most of the economy?
My answer is burn it all down...
Ed mostly avoids talking about LLMs and instead talks about money.
"It is difficult to get a man to understand something, when his salary depends on his not understanding it."
Hard to know of Zitron actually believes what he writes, is trapped in some type of mania, or keeps doing the same bit to pay the bills. Probably a mix of all 3.
Because he‘s not?
There are in fact very few things he got right. I can’t point to 2 things he predicted and came out true.
He once said CoreWeave is going bust. Since then the stock tripled lol.
He said ChatGPT will no longer get more users in 2024.
He said that models haven’t improved significantly since. GPT 4.
He predicted in 2024 that OpenAI will collapse in few months to two years unless it creates a totally new form of AI.
Frankly at this point no one should trust his analysis.
He's a grifter who uses his followers to make money. He tells his followers stuff that he likely doesn't believe in himself but his followers want those things to be true.
To me the main question is what happens after that..
We all move to frozen open source models running on 2nd hand Oracle/Coreweave GPUs? 10 years before someone dares make another training run?
Culturally, do we all collectively sober up once money dries and hallucination are still here? Pendulum swing, AI consideredharmful moment? How to promote healthy use when cognitive surrender is so engrained in us?
What happens if there's a new GPT2 scale (i.e. not astroturf/mass histeria marketing) breakthrough?
I find that modern harnesses mostly get rid of hallucinations: ask your agent to provide quotes and citations and tests for everything, and have another agent cross check that.
Same way we used to do that as humans: I vaguely remember a case or an anecdote from history, but before I use it in a text, I go and look up whether my memory is playing tricks on me.
I mean the companies are delaying the inevitable. We have companies putting limit on the monthly token spend. Just 1% (you see what I did) of the companies can spend fearlessly on AI. In time most of the companies will be burnt out of their funding. I hope AI can make a case to be affordable.
This year a lot of companies got the green light to buy AI subscriptions for 1 year.
The fun begins the next financial year when leaders need to take stock and justify the costs.
Income by implemented features, losses by bugs and savings by personal. Sticks and stones may break even my bones, but ledger words will never hurt me.
Zitron is literally the worst person to raise alarms about the financials of the AI ecosystem because he's so hyberbolic and pollutes his own arguments with nonsense.
Take:
> When somebody decides to build an AI data center, they form a special purpose vehicle (much like a CDO), which then raises debt, in some cases slices it into tranches and, in most cases, sells them to institutional investors, asset managers or banks.
This is just such a weird and wrong comparison. A CDO's assets are other people's debt claims. The same mortgage bond could be split among many CDOs at once, those CDOs could be re-tranched into further CDOs, and thanks to credit default swaps, synthetic CDOs could reference bonds nobody in the deal actually owned. So basically exposure to a fixed pool of mortgages could be manufactured without limit.
A data center SPV's assets are the building, the power interconnect, the GPUs, and the customer contract. If the SPV fails, the loss is limited to what those things are actually worth. There are no multipliers as there are with CDOs.
Later in the post, Zitron even concedes this:
> What differs this from the subprime mortgage crisis is that the systemic risks aren’t driven by derivatives or complex financials but by the sheer scale of costs to build an AI data center, a catastrophic misunderstanding of the AI industry itself and the dangerous lending standards of private credit.
He claims this isn't important:
> When every single debt deal is over $500 million and usually numbering in the billions, we don’t need a vast web of different contracts to create a systemic risk, just clusters of projects that either fail to keep up with their SPVs’ debt or bonds that go unpaid by destitute or defunct data center developers.
But here's the thing: systemic risk isn't a function of how big the losses are. Instead, it's a function of who takes the losses and whether they propagate.
Equity holder losses just get absorbed by equity holders. What happened in 2008, on the other hand, was that the losses hit leveraged intermediaries funding long assets with overnight money, so one firm's distress became another firm's funding withdrawal.
Big deal sizes don't create that type of situation. A $10 billion SPV default is a $10 billion loss distributed across whoever bought the debt.
He brings up Lehman but that's literally the worst example for his argument. Lehman's losses were trivial against its $600 billion balance sheet. It failed because of a funding run. Repo counterparties refused to roll, the clearing banks demanded more collateral and prime brokerage clients pulled their balances. This doesn't happen in an SPV because SPV debt is term debt. It's sized and dated to match the asset. There are no runs on a term loan. When an SPV breaches its DSCR defaults, the lenders take the assets. It's not pretty, but it's contained. It can't spread beyond its own confines and multiply because there is no maturity mismatch, which is what killed Lehman.
A warning: I'm very naive in anything financial.
What happens if the assets collected drop in value as they get repoed? Wouldn't the lender now also be in harms way and in turn have issues financing themselves?
Thanks
> Wouldn't the lender now also be in harms way and in turn have issues financing themselves?
Yes, they could be. There are potential multipliers here, and one failed project could cause other projects to be marked down, even if they're not failing.
But the mechanisms by which $1 trillion in data center debt could become multi-trillion dollar write downs just doesn't exist the same way it did with subprime mortgages.
In the mortgage market, credit default swaps let the same collateral be referenced without limit, CDO-squared structures re-tranched the losses into concentrated wipeouts, and the paper was on the balance sheets of firms funding 30 year assets with overnight repo.
A derivative market for individual SPVs where there's a mismatch between the debt term and the asset just doesn't exist in the data center market.
Pensions funds that are involved have 30 year liabilities and no redemption pressure. Insurers don't have run risk. Banks hold the debt against capital buffers and they have access to the discount window. The riskiest debt holders are levered private credit and semi-liquid retail vehicles, but even in these cases, you're looking at leverage around 2:1, not 30:1.
None of this is to say that an AI bust couldn't cause widespread pain. Just look at how much of US equities are dominated by AI-linked companies. But the issue is that Zitron is just wrong about the mechanisms and magnitudes. He's trying to paint a 2008-like scenario because it's much scarier.
CDOs and SPVs have two things in common:
They're acronyms and they are financial engineering.
Maybe this time it will be different?
A gun and a bomb are both weapons. It doesn't mean they do the same damage.
The problem is the current media landscape/environment rewards hyperbole and drama. Calm narratives of the facts doesn't find much traction.
Yes, but nothing 'current' about that.
Ed Zitron is right about the need for a market correction, but he's fundamentally wrong on AI itself. He'll get his 15 minutes of "see I told you so" when the market corrects, but over the long term he'll be proven wrong.
Ed's primary gripe is that he thinks the business models aren't viable for profitability. But Google's AI infrastructure buildout is already spending less then the depreciation value of the hardware, meaning it's inevitably going to become profitable - at least for Google. Microsoft has since adopted the same approach that Google is using, focusing on faster and more efficient models in order to reduce costs.
Ed won't acknowledge that the paradigm shift for programming and SWE has already happened. He won't acknowledge that roughly 30% of radiology labs in the US and 40% of dental practices have adopted AI.
I personally think Ed is digging a hole he won't be easily able to climb out of.
You seem wildly confident about the market, so I ask.
What market correction are you expecting and by when?
Can you make a falsifiable prediction without moving goal posts and ambiguous timelines?
Or just show us your short position. :)
> Can you make a falsifiable prediction without moving goal posts and ambiguous timelines?
I find this demand dishonest. The inherent nature of bubbles is that you cant predict when they pop. The more they grow, the more damage they cause, but it is impossible to guess what exactly triggers the pop.
So yes, ambiguous timeline is actually correct way to phrase things. Anything else is pure gambling.
I'm a SWE and use AI all day every day. I also acknowledged the very real rollout of AI in other facets of the economy besides SWE. But a market correction is inevitable.
We are in the wild west of AI. No one truly has agentic workflows "down pat" yet. We have no idea which AI workflows, pipelines, architectures, and platforms are going to stick around long term. For all we know a kid in high school can be the one who comes up with a SWE workflow that FAANGs and the rest of the tech sector ends up adopting.
Recent numbers from Google suggest that Ed Zitron might very well be wrong about the data center build-out, which is the single biggest crux of his argument. But even if he's wrong on that, he'll still get his 15 minutes of "see I told you so" (while still being wrong) because even if the data center rollout doesn't implode from assets depreciating too quickly, a correction will still need to happen on the dirt end of the shovels. Even if you don't have too many shovels (which Ed thinks we do), you still need to be digging in the correct patch of ground to strike a gold vein. There's going to be winners and losers in this thing, and the market correction will happen once the biggest investors and hedge funds have correctly identified who the losers are.
What I can't do is provide a timeline. I don't claim to be a prophet. However, if Ed is correct on the hardware buildout, the sheeit will hit the fan once amortization schedules hit critical mass. Otherwise it will be whenever the big money learns who the winners and losers are on the implementation rollout. That can happen at any time. It's perfectly normal and is not a doomsday clock. That's why it's called a market *correction*.
I thought the AI bubble was supposed to pop in two weeks a month ago?
the market can remain irrational longer than I can stay solvent
1. Make a prediction about bubble
2. Get it wrong
3. Excuse yourself by claiming market is irrational
That is the quote of famous financial advisor made in 1986. It keeps being repeated, because bubbles and their pops and irrational markets are frequent enough.
There is nothing controversial about claim that markets are irrational, bubbles pop impossible to predict while bubble being visible. Previous bubbles were the same.
The problem with this type of thinking is that it deliberately ignores the times that bubbles were predicted incorrectly.
Imagine there was a guy. He kept predicting that a war may happen in 1 year. Keeps getting it wrong. In 59 a years a war does happen. He then retroactively justifies his wrong predictions by claiming “but people are irrational and it’s hard to predict”.
Past bubbles I read about were recognized correctly for years before they popped. The pop was hard to predict, the bubble was routinelly recognized. These were not wrong prections, these were correct assesments of the situation. Bubble exists before it pops.
Markets being irrational is again, nothing controversial.
That is history. Your war predicting guy is a made up analogy. And he may just be correct about his assesment of the situation - war is being actively planned and prepared, actors want it and we are all waiting till Trump or Hitler or whoever decides that "now".
Nah, it must have already happened, I saw a video on YouTube that said it was imminent a year ago.
Nobody with a minimum of credibility says that. Everyone says this is clearly a bubble and yet it could go on another couple of years. Nobody is timing it. But if you want to deny the obvious by numbers go ahead
Ed has timed it.
https://x.com/edzitron/status/1817955630784917548
> Newsletter: Based on estimates of their burn rate and historic analyses, I hypothesize that OpenAI will collapse in the next 12-24 months unless it raises more funding than in the history of the valley and creates an entirely new form of AI.
He posted this in 2024 June btw. He certainly doesn’t believe that OpenAI has a new form of AI. So where’s the collapse? Since then the revenue growth was exponential.
I like Ed in general but he’s acting as an influencer, therefore I wouldn’t place him under the broader credibility umbrella.
Zitron's posts reminds me of all the analysts who predicted the collapse of TSLA based on the fundamentals, and recommended shorting it. How does that saying go? "markets can remain irrational a lot longer than you can remain solvent".
There is an immense, coordinated, desperate effort to keep the biggest party in the history of the financial system going, to pour enough fuel into the engine that escape velocity may be achieved against the forces of mathematics.
Funny I should mention TSLA since the stock seems to be in freefall at the moment - https://finance.yahoo.com/markets/stocks/articles/tesla-stoc...