Zitron: The Subprime Datacenter Crisis
wheresyoured.at
wheresyoured.at
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.
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.
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.
My answer is burn it all down...
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.
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?
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.
The fun begins the next financial year when leaders need to take stock and justify the costs.
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?
Thanks
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.
They're acronyms and they are financial engineering.
Maybe this time it will be different?
2. Get it wrong
3. Excuse yourself by claiming market is irrational
There is nothing controversial about claim that markets are irrational, bubbles pop impossible to predict while bubble being visible. Previous bubbles were the same.
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”.
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".
He is not giving investment advice, in any case.
You can’t keep saying bubble will pop and then retract. People catch on.
Zitron's claims are wrong even if the AI bubble ultimately pops. Being directionally right about stuff like this is worse than useless and makes people overcredulous of the other side.
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.
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.
What market correction are you expecting and by when?
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.
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*.
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...