Transcription
The man who saw the 2008 great financial crisis coming before anyone else on Wall Street would listen has just put real money down on the next disaster. And this time his target is the artificial intelligence trade that props up the whole entire market. Michael Burry has escalated from betting against a couple of AI names to shorting the whole entire AI complex and he's calling it a sand castle.
So, is the man with the best crisis call of the century too early again or dead right? So, today we break down the bet he actually placed and the names he's targeting and why he swears this time is just like before the dot-com crash and why a handful of overpriced chip stocks could drag down the savings of people who never even touched them. My name is Nick and this is the Finance Bureau.
Now to understand why this matters, you have to understand how Burry usually operates. For most of his career, his big bets have arrived through the regulatory paper trail, the 13F filings that every large fund has to submit to the SEC. And in his last confirmed filing for the third quarter of 2025, the positions were already aggressive. He held roughly 5 million put options against Palantir, a notional value of around $912 million. He held about 1 million puts against Nvidia, worth roughly $187 million in notional terms.
Now, without getting too technical, a put contract is just an instrument that pays out when the stock price falls below another certain price, so-called the strike price. So all of Michael Burry's put positions was a straightforward bearish stack against the entire AI story. Combined, the notional value of those trades came to about $1.1 billion. Now to be clear, notional isn't a cash position at risk. The actual premium he paid on the Palantir position was reportedly at roughly 1% of that headline figure.
But then something changed. In November 2025, Burry's fund, Scion Asset Management, deregistered with the SEC after falling below the reporting threshold. Put simply, the paper trail went dark. No more 13Fs, no more independently verified positions. Instead, he launched a paid newsletter called Cassandra Unchained, and that is now where he discloses his trades. So everything that follows comes from Burry's own word, not a regulator stamp. And that's an important caveat, and we'll come back to it in a bit because what he disclosed next was not a whisper.
At the end of June 2026, in a post titled simply "Trading Post," Burry laid out a basket of direct short positions. He shorted Nvidia at $198. He shorted Applied Materials at $729. He shorted Tesla at $416, a name he actually dubbed the lord of what he called a tragic tier of stocks. He shorted Caterpillar of all names at $1,060, his first ever short on a company he used to own, targeting its as a proxy for the data center construction boom. And a couple of days later, he added a direct short on Micron at $1,051. He even rolled his existing bets against the semiconductor index further out, pushing the expiry to March 2027 and raising the strike prices to $400. Longerdated, higher conviction.
The word he reserved for Palantir was blunt. "A sand castle," he said, "supported only by the AI applications narrative priced for flawless multi-year execution that simply won't hold." He values it at roughly 16 times his own estimate of what it's actually worth. So this is conviction named and priced.
But the question underneath all of this is why now? And here Burry reaches for the one comparison guaranteed to make investors flinch. The year 2000. In May 2026, he said the market was feeling a lot like the last months of the 1999 to 2000 bubble. And he has a specific number he keeps pointing to: the Philadelphia Semiconductor Index. So the benchmark for chip stocks, which was trading around 65% above its 200-day moving average. Now, the 200-day moving average is just the average price over the last 200 trading days, which is a slow line that smooths out the noise. When price runs far above that line, it means the rally has stretched itself violently. And Burry's claim is stark. The last time chips were stretched this far above the trend was the dot-com peak.
Now for context, the tradable proxy most people can actually buy, the iShares Semiconductor ETF, closed at $566, about 50% above its 200-day moving average. So even the watered-down version everyone seems to hold is running hot, and Burry's index figure is hotter still. And his warning is that stretches like this have historically been followed by declines of 30% or more. And his favorite parallel is Cisco. In 2000, Cisco was the picks and shovels king of the internet. Exactly the way Nvidia is the picks and shovels king of AI today. Cisco fell around 80% to 90% after the crash and remarkably only just a few months ago recovered those all-time highs. The company survived, but the shareholders who bought the top the last time around had to wait for a very long time.
South Korea's announcement of an enormous semiconductor spending wave, roughly $500 billion from Samsung and SK Hynix combined. His words, "I see that as the beginning of the end." And he frames the whole rally as psychology, not fundamentals. A fear of missing out, greater fool theory, and what he calls public commitment bias. Everyone's in, so nobody can admit they're wrong. And that begs the question of how an ordinary investor is supposed to cut through the noise when the whole market has decided to look the other way. And that's exactly why we built the Finance Bureau newsletter. It's completely free and we break down the markets, macro, and monetary policy so you never miss what truly matters. Just click the link in the description or scan this QR code on the left of your screen.
And now back to Burry's evidence. Now, Burry's real case rests on two mechanical red flags. And the first is the plumbing of the AI boom itself. He calls it circular financing. And here's how it works. Nvidia invests billions of dollars into AI startups and cloud providers. These companies then turn around and spend that same money on buying Nvidia chips. So Nvidia's investment comes back as Nvidia's revenue, and the whole chain reports explosive growth. Put simply, the company is helping to fund its own customers. And the numbers behind this aren't small. In the first four months of 2026 alone, Nvidia committed over $40 billion to AI equity investments, including a reported $30 billion stake in OpenAI. And one note that Burry keeps circling on is the NeoCloud, which are basically companies like CoreWeave that borrow against Nvidia's hardware. CoreWeave secured debt facilities worth around $8.5 billion collateralized by the very chips the boom depends on. And Burry mapped out one financing structure and tagged the whole thing "a fugazi." He described deals running through 8 to 12 Byzantine steps, weaving in insurance money, reinsurance, and private lending, some of it sitting off the balance sheet entirely. And his argument is that these arrangements funnel ordinary people's retirement and annuity money into inflated demand for chips.
But the more concerning thing here is that Burry isn't alone. Legal scholars have flagged antitrust concerns over these interlocking dependencies, and analysts estimate OpenAI's combined chip and cloud commitments could top $1 trillion. The scale is real, even if you just ignore Burry entirely.
The second red flag, though, is subtler, and it's about how these giants report their profits. Burry argues that the hyperscalers, the Microsofts, Metas, and Oracles of the world, are stretching how long they pretend their chips will last. On their books, they depreciate that hardware over 5 to 6 years, but Nvidia releases a faster chip every single year. So, the real useful life, Burry says, is closer to 2 to 3 years. Stretch the depreciation schedule and you spread the cost thinner, which makes today's profits look bigger than they actually are. His estimate for the size of that gap: roughly $176 billion in understated expenses across the industry between 2026 and 2028. He called it one of the more common frauds of the modern era. And some analysts extending his thesis argue that Oracle and Meta could be among the most exposed to understated depreciation, though precise overstatement percentages aren't independently verified. And some estimates building on this thesis suggest reported industry profits could be meaningfully inflated, though the exact scale is also disputed. And it sits on top of a spending gap that even the bulls admit is enormous. Hyperscaler capital spending is projected to hit around $725 billion this year. And the revenues it's actually generating. Estimates vary widely, but they remain far below current spending levels according to several analysts.
And that's the case for the prosecution. But I want to be fair because Burry has a track record that cuts both ways. And the counterargument here is genuinely strong. Let's start with the man himself. Burry is famous for being early, and early can look identical to wrong for years. He shorted Tesla repeatedly since 2020, and those bets have punished his portfolio. Elon Musk has publicly mocked him as a "broken clock." He held bearish bets against the broad market through 2023 while stocks kept hitting all-time highs. And on Palantir, the stock he called a sand castle, he was criticized for being early as it kept climbing before it eventually fell by around 44% from its peak. He then covered half that position at $117, which was a partial win, but not a clean move.
But there is a bit of nuance here, guys, because in June 2026, Burry disclosed a bullish, long-dated call on Microsoft, betting rises through 2028. He's not against all of big tech, and the critics have been loud. Palantir executives have publicly dismissed short-seller criticism of the company's valuation. Analysts noted that Palantir's revenue growth has accelerated sharply in recent quarters, which is not exactly what a sand castle usually does.
But the core rebuttal is the one that matters most. Unlike the dot-com era, these companies actually make money. Nvidia trades at a price-to-earnings ratio of around 29.7. Oracle sits near 23.6. These are expensive, but they are not insane. The Magnificent 7 collectively run net margins above 25% versus roughly 13% for the wider S&P 500. Microsoft alone reports a $37 billion annual run rate from AI revenue. The multiples are real, unlike what we saw in the dot-com era. The honest picture is a bit split. Some names are expensive but profitable. Others, like Tesla at a P/E around 327 and Palantir near 135, are genuinely frothy. Treating an AI stock as one single risk is lazy.
But even the strongest bull case carries a bit of an asterisk, and it's called the Cisco caveat. Cisco in 2000 was profitable. It was a real category leader with real earnings, and its stock still fell around 85% when the valuation reset arrived. Profitability doesn't mean you are immune from a repricing. It just means the company survives while your money doesn't. So even if Burry is wrong about fraud, the valuation risk is stubbornly mathematically real.
Which brings us to the part that actually touches your life. Because you might be sitting there thinking, well, this doesn't really affect me. You don't gamble on chip stocks. You own a boring index fund. You have a pension you never actually look at. And that is exactly the trap because the S&P 500 is a market-cap-weighted index, which means the biggest companies take up the biggest slice of that index. And right now, a handful of AI names dominate it. The Magnificent 7 makes up roughly a third of the entire S&P 500. The top 10 stocks account for roughly 40% of the index value. But here's the kicker. At the 2000 dot-com peak, that figure was about 27%. We are more concentrated now than we were the last time this ended in tears. The technology sector share of the S&P 500 is at or near record highs, and by some measures exceeds the dot-com era peak, and the market is being carried by fewer and fewer shoulders. In several recent rallies this spring, a small handful of mega-cap stocks accounted for the large majority of the index's gains.
So when you buy a standard S&P 500 fund, you think you're buying 500 companies. In reality, nearly half of your money is riding on a dozen AI-linked names. And let me tell you something else. If you hold a target-date retirement fund, you also own them. If you have a workplace pension tracked in the index, you own them. If your savings sit in the most popular ETF on the planet, you own them. And you probably never chose to.
And that's the mechanism Burry is really warning about. A crash in six or seven names doesn't stay contained in six or seven names. It drags the whole index down, and everyone tied to that index goes with it. Which is the quiet asymmetry at the heart of all of this. The people running these companies, the insiders, the funds with the deepest pockets, they can read the concentration chart. They can hedge, they can rotate and sell into strength. And some, like Burry, are already positioned for the downside with money most of us will never have. The everyday saver holding the index doesn't get that luxury. They're told diversification keeps them safe while the index they trust has become one enormous bet on a single theme.
If the AI trade keeps compounding, the insiders and the ordinary holders both win, and Burry looks again like a broken clock. But if it snaps, the sophisticated money is already out the door, and it's the pension holder who supplies the exit liquidity. Same market, two very different outcomes decided by who saw the concentration coming. And that's the loop that closes here. Burry called 2008 not because he predicted the future, but because he read the plumbing that everyone else just ignored.
Whether he's early again or just right on time, the main takeaway here should be awareness. You don't have to agree with his thesis on the sand castle to check how much of your own portfolio is standing on the same six names. Watch the concentration in your own funds. Watch where the hyperscaler spending ever turns into durable revenue. Watch whether the market's gains broaden out or keep narrowing into a handful of tickers. And separate the expensive but profitable from the genuinely frothy rather than treating AI as one big monolith. There is even a simple tool for the concentration worry. An equal-weighted version of the same index holds the same 500 companies but cuts the Magnificent 7's influence from around a third down to roughly 1.4%. Same companies but a fraction of the single-theme risk. But at the end of the day, guys, the best protection is knowing what you actually own before someone else decides its price for you.
Well, that's all we have time for today. But what do you think? Is Burry the broken clock that the bulls say he is, about to be steamrolled by record earnings once again? Or is this genuinely 2000 all over again with the whole market unknowingly leveraged to a sand castle? Let me know your thoughts in the comments down below. And if you want to understand how this same concentration risk has warped the housing market too, you can check out our video on that right over here. And that's all from me today. As always, thank you very much for watching, and I'll see you in the next video. This is Nick signing.