원문 보기
In 2005, a doctor with a glass eye pits the entire American housing market against each other. I think I know for sure : some of them thought I was crazy. Three years later, the market collapses. He brings in hundreds of millions of dollars for his investors, and Hollywood turns him into Christian Bale.
20 years later, Michael Bury is betting again. This time—against the most valuable company on Earth, Nvidia, the first company in history to surpass the $3 trillion mark in value. And this time he's not just betting against one company. He is betting against a vicious circle. Nvidia is investing up to $ 100 billion in OpenAI.
OpenAI spends them on Nvidia, Oracle, AMD, CoreWeave chips . Critics call it the same money going in circles. Companies call it building the future. The CEO of Palantir, one of the two companies he's betting against, goes live. Hell, the two companies he's shorting are the ones making all the money, which is really weird.
Like, the idea that chips and ontology are something that needs to be "shortened." This...this is madness. Nvidia's CEO is calmer. I don't believe we are in an AI bubble. So, who is right? The person who predicted 2008, or the company that posts record profits quarter after quarter? By the end of this video, you'll see one number that doesn't match, and you'll have to decide if it's already in your retirement account.
Michael Bury is a doctor by training. He lost an eye to cancer as a child, trained as a neurologist, and then quit to run a small investment fund, Scion. In 2005, he does something that almost no one on Wall Street cares about. He reads mortgage bonds, every single one . Yes. Aha. Nobody reads them.
They are read only by the lawyers who created them. I don't think they even know what they created. He finds loans that cannot be repaid and convinces banks to sell him insurance against them. We bought, in essence, " short" $8.4 billion in credit default swaps related to mortgages or financial companies.
In 2008, the housing market crashes. From 2000 to 2008, Scion generated nearly 490% of its net profit after commissions were deducted . And how did they thank him? I think a lot of customers were just glad it was over, you know. Then the book, then the movie "The Downgrade Game." The person who was right when everyone else was wrong.
But here's what the movie won't tell you. After 2008, Bury often makes mistakes. He is betting against Tesla in 2020. Tesla continues to grow. In January 2023, he tweeted one word: " Sell." Two months later: "I was wrong to say sell." So who is he really? Cassandra, the prophet no one listens to, or the person who keeps shouting "fire" until one day it happens?
October 2025. After almost two years of silence, Berry returns online under the new name " Cassandra Unleashed" with a mystery. In a few days, his fund's quarterly report becomes public. Put options on Palantir and Nvidia, controlling shares worth over a billion dollars. Together, this is about 80% of the portfolio he reports on.
To be precise, this billion is the value of the shares controlled by the options, not the money he actually paid. But this message cannot be ignored. A week later he does something even stranger. He closes the fund. " I have come to the unfortunate conclusion that I must close the fund." His explanation consists of one sentence.
Then he takes on accounting . Nvidia releases a new generation of chips every year or two. But the biggest buyers— Microsoft, Google, Amazon, Meta, Oracle— depreciate their servers over five or six years. Berry estimates that this could overstate their combined profits by $176 billion between 2026 and 2028 .
This is his assessment, not an audit. And Nvidia is doing something it almost never does. She answers. A seven-page memo to Wall Street analysts, with one name at the top. Berry's answer. So what exactly is he betting against? To understand this, you need to follow the money in a circle. September 2025.
Oracle signs a deal to sell OpenAI computing power worth about $300 billion over five years. 11 days later, Nvidia announces it will invest up to $100 billion in OpenAI . And OpenAI plans to fill data centers with at least 10 gigawatts of Nvidia systems. And Oracle, when building these data centers, also buys Nvidia chips.
Then AMD. OpenAI is committing to purchase 6 gigawatts of AMD chips, and AMD is granting OpenAI warrants to purchase up to 10% of the company , which vest only after OpenAI meets its deployment goals and only if AMD shares continue to rise . Then Broadcom, then Coreweave, a cloud company partly owned by Nvidia, whose unused capacity Nvidia has promised to buy up if no one else does.
Bloomberg draws this diagram. The supplier buys a portion of its customer. The customer uses the money to buy from the supplier. Revenue on one side , investments on the other. Over a trillion dollars in commitments in a matter of weeks. Back in January 2025, at the Stargate presentation at the White House, Sun made a promise.
And you told me, "Oh, Masa, aim for 200." Now I'm back with 500. In November 2025, SoftBank sells all of its Nvidia shares to invest more money in OpenAI. And this June, Sun says this. I think it's 10, maybe 50 times bigger than the dot-com crash. And Steve Iceman, another real-life investor from The Short Game, played by Steve Carell, sees a weak link.
So if something happens to one of these two companies, the chain will really break. And of these two, OpenAI is the weaker company. So, let's look at this link. Here are the numbers that don't add up. By the end of 2025, OpenAI announces plans to reach revenue of about $20 billion per year. The cost of computing, according to Sam Altman himself, is about $1.4 trillion over eight years.
20 billion in revenue, 1.4 trillion in liabilities. Reports indicate that OpenAI expects to spend about $115 billion in cash by 2029. When one of OpenAI's investors, Brad Gerstner, raises these numbers with Altman on his podcast, this is how the conversation ends . And if you want to sell your shares, I will find you a buyer.
I have enough. Altman's argument: the bigger risk is not having enough computing power, not too much. And it's not just OpenAI. Microsoft, Google, Amazon, and Meta plan to spend about $730 billion on infrastructure in 2026 alone. Some of these costs are taken off the balance sheet. Meta's massive data center in Louisiana is financed through a $27 billion joint venture that is not reflected in Meta's financial statements.
And Oracle, the company that signed the largest contract, is seeing the cost of insuring its debt against default exceed 2008 levels . When Mark Cuban was asked about the profitability needed to justify these numbers, he didn't hesitate. Regarding the profitability needed for these numbers to make sense.
They will never get it. So why isn't everyone selling? Because Nvidia's money is real. $57 billion in revenue in one quarter, then $68 billion. Profitability over 70%. Most dot-com companies didn't make a cent. I believe the demand for computing has increased a million-fold in the last 2 years. Your data center is now a token generation factory.
But this is a place where when companies and partners take bold steps, they will be rewarded. Even those who talk about a "bubble" don't call it fake. Jeff Bezos. This is a kind of industrial " bubble" as opposed to financial "bubbles." An industrial " bubble" leaves something behind when it bursts.
Railways, fiber optics, technology remain, even when investors leave. And there's this. Sam Altman himself at dinner with reporters. It is both rational, but at such moments there are also elements of irrationality. No company will be safe, including us. Not everyone who builds it generally agrees that a "bubble" exists.
Huang and Su categorically reject this word. But even those who use it, like Altman and Pichai, see it as a risk worth taking, not a reason to stop. We've seen this movie before. March 2000 , Cisco, a company that sold equipment to build the Internet, briefly became the most expensive company in the world.
Its customers, new internet and phone companies , need billions to buy this equipment. That's why Cisco, Lucent, and Nortel lend them these funds. By the end of 2000, nine equipment manufacturers had lent their own customers more than $25 billion . Then the customers were unable to pay. Between 2000 and 2003, dozens of them went bankrupt.
Cisco shares fell nearly 90%. Cisco's profits grew exponentially, but it wasn't until December 2025, a quarter-century after the crash, that its stock finally returned above its 2000 peak, buoyed by the same AI boom that is now in question. So is it the same? Not quite. Today, Nvidia's customers mostly pay in cash, and they are some of the richest companies on Earth.
This is the strongest argument against Burry. His answer this month is another history lesson. The computer leasing boom of the 1960s, when companies bet that mainframes would last for years. Then IBM released a machine that made the old leases worthless a few years earlier. So where are we in October 2026?
Nvidia continues to report record quarters, and in early June, the chipmaker's shares lost more than a trillion dollars in value in a single trading session, then clawed some of it back for weeks. Burry sells put options expiring in December of this year in September and moves his bet to 2027. Did he win or lose at this point?
There are no public figures. Anyone who tells you otherwise is just guessing. Here's what is known for sure. In October 2025, the IMF and the Bank of England warned that just five companies would account for about 30% of the S&P 500 index—the highest concentration in 50 years. If you have an index fund, pension, or 401K, you're not watching this circle from the sidelines.
You are already inside it. In 2008, they thought he had gone crazy. In 2023, he advised selling, and he was wrong. Now he says it again. The question is not whether
Comments 0