Michael Burried

Start with the book itself. On August 10 and 11 Burry disclosed he added to his "three largest shorts": $NVDA puts expiring December 2026 and June 2027 with strikes in the low $100s, $PLTR puts expiring December 2026 and March 2027 also struck in the low $100s plus more shares shorted outright at $175, and increased shorts in $ORCL at $145, $CAT at $844, and $SOXX at $533. He also added to a long position in $MOH, a healthcare insurer down about 54% from its high. He framed the setup as a replay of 2005, the year before the housing crash actually broke, comparing today's AI capex boom to $GS quietly offloading a $3 billion mortgage position just before "the floodgates opened" in 2006.

Here's the part getting skipped in the recap. $PLTR reported one of the strongest quarters in its history on August 3: revenue of $1.935 billion, up 93% year over year and beating consensus by roughly $123 million, US commercial revenue up 149% to $764 million, GAAP net income of $1.06 billion versus $327 million a year earlier, and full year guidance raised from $7.65 billion to $8.15 billion. The stock jumped roughly 30% in the days that followed, and per a Bloomberg/S3 Partners tally cited by TechTimes, short sellers lost about $3 billion in paper gains in that single session. This week, $PLTR is trading around $170 to $175. Burry's new puts, struck in the low $100s, need the stock down close to 40% from here just to reach the money, and that's before his separate claim that the stock is intrinsically worth under $1 is even in play.

The freshest example undercuts the "vindicated bear" framing even harder. On August 6 Burry disclosed a short in $NBIS at $211.77, a position he called larger than the $ORCL short he opened the same day at $144.63, choosing a direct equity short over puts because implied volatility exceeded 100%. Nebius reported second-quarter results before the open on August 12: revenue of $582.3 million, up 454% year over year and above estimates, an adjusted per-share loss of $0.12 versus a forecast loss near $0.82, reaffirmed full-year guidance, and a contracted-power target raised to 5 gigawatts. The stock jumped as much as 34% to around $259, roughly 22% above the price he disclosed shorting at. Burry's response to a losing trade wasn't to cover, it was to add. That same day he disclosed pressing the Nebius short further at $247, alongside fresh adds to his $MU short at $924, his $ORCL short at $152, and his $SOXX puts, calling Nebius "what the top of a boom looks like" and pointing to the company stretching server depreciation from four to five years as evidence the real decay rate is still being hidden. It is the single clearest data point of the week for the piece's actual argument: the size of the shorts keeps growing in the same week the earnings keep beating.

$NVDA is the cleaner example of a leg that simply hasn't worked yet. The stock closed near $223 on August 12, market cap around $5.4 to $5.5 trillion, inside its 52-week range of $164 to $237 and not far from its all-time high. Burry's puts, struck in the low $100s, need something close to a 50% collapse just to reach the money. 24/7 Wall St's direct read on his own disclosures this month says it plainly: he's claimed profits on his $PLTR, $MU, $TSLA, $CAT, and $AMAT shorts along with $SOXX, but "his bet against Nvidia has not yielded a profit." The same piece adds that Burry has never publicly closed any of these positions, which makes the claimed gains elsewhere impossible to independently verify either. And on the depreciation logic underneath the whole NVDA short, $CRWV disclosed on its own earnings call this month that it has A100 chip contracts, on hardware that first shipped in 2020, signed through 2029. That's a direct data point against the idea that this generation of chips goes economically dead early.

Zoom out further and the pattern holds up before this AI trade even started. An independent tally by Yavuz Akbay of Burry's major public calls between 2017 and 2023, separate from the current AI short, found roughly 71% of 14 calls were wrong, with the clear exception being a February 2021 call on rising inflation. His fame still rests almost entirely on 2008. Since then: a bitcoin-to-housing comparison in March 2021, followed three months later by a warning of what he called the worst crash in history, on a timeline that never arrived. He's conceded this himself: "I am now a meme for the number of times I have called a crash," he wrote. "I have become the boy who cried wolf."

He doesn't fully dispute the pattern, he reframes it. When an X account laid out his misses in July and labeled him "Captain Broken Clock," Burry's entire reply was "nice graphic, but nothing on it is true." Elsewhere he's argued that judging a short thesis years after the fact misreads how short positioning works: "you really think any short seller holds those positions for 5 or 10 years?" Fair point on trade duration. It doesn't change how many of the underlying predictions, on their own stated terms, simply didn't happen on the timeline he gave them.

And there's a business model sitting underneath all of this that rarely makes the recap. Cassandra Unchained, the Substack he launched the same month he closed Scion, crossed 300,044 subscribers by mid-July, just 231 days after launch, across 212 countries, at $39 a month or $379 a year, a theoretical $113.7 million a year if every subscriber paid the full annual rate. Even after Substack's standard cut, that's a business that requires nothing to go right on any specific strike price by any specific date. Being the man who called 2008 is now, on its own, a more durable revenue stream than being right about $PLTR going under $1, and newsletter growth doesn't care which one turns out true.

None of this means his underlying depreciation argument on AI infrastructure is wrong on the mechanics, it's a genuine accounting question worth tracking on its own terms, separate from whether any specific strike price gets hit on any specific date.

He might still be right on the mechanics years from now, the way 2008 eventually arrived years after he first bet against subprime. But right now more of his own book is underwater than the "vindicated bear" recap admits, most of his career since 2008 reads the opposite way by an independent count of his own calls, and none of it changes the subscriber count either way. Watch the strikes, not the newsletter, if you actually want to know which trade is winning.