13F Portfolio Analysis — Michael Burry | A Portfolio for the AI Bubble Era
Michael Burry warns of an AI bubble. How did he build his own portfolio?
The ‘AI bubble’ debate is heating up again, with Michael Burry, the real-life hero of The Big Short, at its center.
In late September, Burry said the AI bubble may burst sooner than he first thought (CNBC, Sept. 28).
So we set out to analyze what strategy Burry, the leading voice of the AI-bubble camp, actually used to build his portfolio.
Burry’s firm, Scion Asset Management, deregistered with the SEC in November 2025, so its last official 13F covers Q3 2025.
We split 12 quarters of Scion’s 13F filings (Q4 2022 to Q3 2025) into two portfolios and ran them through DIVA Quantizer.
· Long book — what he bought, as stock or call options
· Short book — what he bet against with put options
How the two portfolios move against each other reveals Burry’s view of the market, and 12 quarters in a row show how that view changed. Finally, we look at how ‘it may burst sooner’ shows up in his portfolio today.
1. Burry’s two portfolios — the long book and the short book
Scion’s last 13F (Q3 2025) totaled $1.38B, and 79.5% of it was Palantir and NVIDIA put options.
All eight lines of the last 13F
| Holding | Type | 13F value | Weight | Side |
|---|---|---|---|---|
Palantir PLTR | Put | $912M | 66.0% | Short |
NVIDIA NVDA | Put | $187M | 13.5% | Short |
Pfizer PFE | Call | $153M | 11.1% | Long |
Halliburton HAL | Call | $62M | 4.5% | Long |
Molina Healthcare MOH | Stock | $24M | 1.7% | Long |
Lululemon LULU | Stock | $18M | 1.3% | Long |
Sallie Mae SLM | Stock | $13M | 1.0% | Long |
Bruker (preferred) | Stock | $13M | 1.0% | Excluded (no price data) |
The value a 13F lists for an option is notional: the number of shares it covers × the quarter-end price. The Palantir put is 5 million shares × the Sept. 30 close of $182.42 = $912M.
Right after the filing, Burry said on X that he actually paid about $9.2 million for these puts (as reported), about 1% of the 13F value. They let him sell Palantir at $50 a share in 2027, 72.6% below the Sept. 30 price. However high the stock goes, he can lose no more than he paid.
Here is how they went into DIVA. The long book is stocks and calls (calls at their 13F value, as if he held the stock). DIVA doesn’t accept negative (short) weights, so for the short book we entered the stocks under the puts, weighted by 13F value. DIVA shows what kind of stocks these are; it does not compute option payoffs.
The two portfolios in DIVA
| Metric | Long book | Short book |
|---|---|---|
| Holdings (by 13F value) | Pfizer 57% Halliburton 23% Molina Healthcare 9% Lululemon 7% Sallie Mae 5% | Palantir 83% NVIDIA 17% |
| Annual volatility | 19.9% | 90.8% |
| Market beta | 0.87 | 2.15 |
| Annual alpha | −12.2% | +87.8% |
| Explained (R²) | 62.4% | 38.6% |
| DIVA type (FF5) | The Human Index / The Discerning Vault Keeper | The Rodeo Rider |
The long book is 80% Pfizer and Halliburton calls. With a market beta of 0.87 it moves a little less than the market; its DIVA FF5 type is ‘The Human Index / The Discerning Vault Keeper.’
The short book is 83% Palantir by weight and 96.4% by risk contribution — in effect, one stock. It moves more than twice as much as the market (market beta 2.15), and it earned 87.8% a year that the five factors can’t explain (alpha). DIVA calls it ‘The Rodeo Rider.’
Explained (R²) is the share of the moves that the five factors account for. The short book’s is only 38.6%: it moved on Palantir’s own story more than on the factors. We will come back to this number in Sections 3 and 4.
2. Two portfolios that move apart — the ‘two markets’ Burry sees
Five long + two short stocks
The two portfolios are opposites in almost every row. So is the short book insurance for the long book?
Insurance would have to move with what it protects: when the stocks he owns fall, the puts should pay off. We put all seven stocks on one DIVA correlation map.
Palantir’s correlations with the five long-book stocks run from 0.04 to 0.22, and NVIDIA’s from −0.11 to 0.24. The two portfolios’ monthly returns have a correlation of 0.21: they moved almost independently. The short book was no insurance for the long book.
Then what did each move with? Over the same 60 months the long book tracked the US value ETF (IWD) at 0.69 versus 0.39 for the growth ETF (IWF), while the short book tracked the Nasdaq-100 ETF (QQQ) at 0.53 versus 0.33 for value. DIVA’s factor betas make it sharper.
Factor betas of the two portfolios
The two portfolios mirror each other. The long book, with a profitability beta of +0.52 and an investment beta of +0.64, leans toward profitable companies that invest cautiously. The short book, with a value beta of −2.35 and a profitability beta of −2.93, leans toward growth stocks priced on expectations more than profits.
Burry’s position isn’t a bet that the market falls. He sees two worlds inside the market that move separately — profitable old-economy companies and AI companies priced on expectations — and he bought one and bet against the other.
That makes it not a hedge but two independent bets. If AI collapses, the long book barely notices; if AI keeps rising, the long book won’t make up the loss.
By 13F value the short book is far larger, but the money actually paid was about $9.2 million for the Palantir puts. A 13F doesn’t show the options’ real exposure (delta), so this piece does not compute the combined volatility or ‘net exposure’ of the two portfolios.
3. Twelve quarters — from the whole market to one AI stock
Burry wasn’t always betting against AI. Puts appear in 5 of the 12 quarters. For each, we ran both portfolios through DIVA on the 60 months to that quarter-end — only data anyone could see then — and computed their correlation from the same data.
The five quarters with puts
| Qtr | Short book | Corr. with long book | Explained (R²) | Annual alpha |
|---|---|---|---|---|
| 23Q2 | SPY and QQQ | 0.79 | 99% | +1.4%* |
| 23Q3 | SOXX and Booking | 0.76 | 76% | +12.7% |
| 24Q3 | Alibaba, Baidu, JD.com (hedge) | 0.99 | 11% | — |
| 25Q1 | NVIDIA + five Chinese firms | 0.35 | 47% | +31.4% |
| 25Q3 | Palantir and NVIDIA | 0.21 | 39% | +87.8% |
In 2023 his puts were the market itself. The Q2 puts on S&P 500 and Nasdaq-100 ETFs had an explained share of 99% and a correlation of 0.79 with the long book. The Q3 semiconductor ETF (SOXX) puts were similar, at 0.76. When the market fell, his long book would fall too and the puts would pay: a position saying ‘the whole market is expensive.’
In 2025 the puts moved to AI stocks. The correlation with the long book fell from 0.35 to 0.21, and the explained share from 47% to 39%. As the unexplained part grew, so did the target’s alpha: +1.4%, then +12.7%, +31.4% and finally +87.8%.
What a real hedge looks like shows in Q3 2024: puts on the same Chinese stocks he held, with a correlation of 0.99. The last 13F’s correlation (0.21) is the opposite.
The long book changed too. Its market beta was between 0.94 and 1.16 through Q1 2024 and between 0.37 and 0.87 afterward (excluding Q1 2025, when only Estée Lauder was left). As the puts narrowed onto AI, the long book filled up with stocks that move less with the market.
In two years Burry’s puts narrowed from a warning about the whole market to a bet against one AI stock. The target was always whatever had risen more than the factors could explain.
Even Burry the ‘AI bubble’ bear held Alphabet, Amazon and Oracle stock at the end of 2023, and calls on Meta and ASML in Q2 2025. Over 12 quarters, 92 securities appeared in his 13F, and on average 60% of each quarter’s 13F value was in names absent the quarter before. What he bet against was less the technology than its price.
4. And now — ‘it may burst sooner than I thought’
Palantir fell 41.2% from its 13F-date price by June 25, 2026, then climbed back to $209 on Oct. 9, +14.6%. Even its low of $107 was far from the $50 strike (share prices, not option profits or losses).
A year on, the Palantir bet hadn’t paid. Yet in late September Burry said the bubble may burst sooner than he thought. Since Scion deregistered he has disclosed trades in his paid newsletter, ‘Cassandra Unchained.’ Here is what was reported in September.
· Reported Sept. 10: sold December 2026 NVIDIA and Palantir puts, kept 2027 Palantir and QQQ puts. Largest longs: Lululemon, Molina, MercadoLibre; largest shorts: Oracle, Palantir, Nebius
· Reported Sept. 22: added to shorts in Nebius, Micron, Palantir and the semiconductor ETF (SOXX)
· Reported Oct. 4: in late September, swapped the Micron and Nebius short sales for puts expiring June 2027
The weights weren’t disclosed, so we assumed equal amounts and ran them through DIVA (60 months, October 2021 to September 2026). The short book is Micron, SOXX, Palantir, NVIDIA and Oracle (Nebius, relisted in October 2024, has too short a history); the long book is Lululemon, Molina Healthcare and MercadoLibre.
Today: three long + five short stocks
The short book has become one block. SOXX moves with Micron (0.77) and NVIDIA (0.67); only Palantir stands a little apart (0.01 with Micron). The long book and short book have a correlation of 0.25: still moving separately, as a year earlier.
How the short book changed in a year
| Metric | Q3 2025 (13F) | Now (Sept. 2026) |
|---|---|---|
| Stocks | Palantir and NVIDIA | Micron, SOXX, Palantir, NVIDIA and Oracle |
| Largest risk contribution | Palantir 96.4% | Palantir 24.4% |
| Market beta | 2.15 | 1.62 |
| Explained (R²) | 38.6% | 75.5% |
| Corr. with Nasdaq-100 (QQQ) | 0.53 | 0.88 |
| Corr. with long book | 0.21 | 0.25 |
| Annual alpha | +87.8% | +36.9% |
| DIVA type (FF5) | The Rodeo Rider | The Blockbuster Director |
‘Sooner’ shows up in the portfolio in two ways.
First, the net got wider. A year ago the short book was a one-stock bet, with 96.4% of its risk from Palantir. Now five stocks split the risk 15% to 24% each, the factors explain 75.5% of its moves, and it tracks the Nasdaq-100 at 0.88. It is no longer a bet on Palantir’s story but on the whole AI investment wave — memory, chips and cloud — turning together.
Second, he set a deadline. That lies in the options’ structure, which DIVA can’t see, so it gets its own box.
A short sale has no expiry, but its loss grows as the stock rises. A put caps the loss, but the stock has to fall by a set date. As reported, Burry turned his Micron and Nebius bets into ones that must pay off by June 2027 — not ‘someday’ but within about nine months.
The targets he added were also the year’s biggest risers (as of Sept. 30, 2026): Micron +537.6%, SOXX +110.4%, Nebius +110.1%. On the long book, Lululemon, MercadoLibre and Molina sit 22% to 55% below their 52-week highs.
The long book hasn’t changed: Lululemon and Molina, longs in the last 13F, are still his largest longs a year later. A wider net, a deadline, and a long book that moves on its own: that is the mark ‘it may burst sooner’ left on his portfolio.
Wrapping up
Burry warns of an AI bubble, so how did he build a portfolio in the middle of it? Opened with DIVA, his 13Fs come down to three things.
· Two bets that move apart — he bought profitable companies the market had turned away from and bet against AI companies priced on expectations. The two had a correlation of 0.21: not a hedge.
· The target is whatever rose most — the whole market in 2023, Palantir in 2025, and now memory and chips. He bet against AI’s price more than AI as a technology.
· Fix the possible loss — unlike the large notional in the 13F, the money at stake is the puts’ premium; and after saying the bubble may burst sooner, he widened the net and added a deadline.
No one knows when a bubble will burst. Rather than staking everything on the timing, Burry’s portfolio fixes what he loses if he is wrong and waits to be right. Whether he will be is still unknown.
In a portfolio for the AI bubble era, the question isn’t whether it’s a bubble but what you buy, what you bet against, how the two move together, and how much you lose if you’re wrong.
Try entering the 13F holdings of an investor you follow into DIVA Quantizer. Enter what they bought and what they bet against separately, and the market view hidden in a weights table starts to show in the correlation map and factor betas.
① A 13F shows only US-listed stocks and options at quarter-end. Short sales of stock, cash, bonds and foreign-listed shares don’t appear, and option strikes, expiries and purchase prices are unknown.
② Calls and puts entered the two portfolios at 13F (notional) value. That differs from their real exposure (delta) and the money paid; the DIVA results describe owning the stocks. The Palantir put’s premium and strike come from reports of Burry’s own public statements.
③ The Q3 2025 analysis excludes the Bruker preferred shares (1.0% of the 13F), which lack price data, and uses 60 months (Oct 2020 to Sep 2025). The quarterly analysis uses the 60 months to each quarter-end and leaves out stocks with short histories or no price data from the provider (yfinance) (at least 74.8% of the long book’s value is covered in each quarter with puts). Entering the same stocks into DIVA today uses the latest six years, so the figures may differ.
④ Correlations between the two portfolios and with ETFs were computed separately from the same monthly returns and windows DIVA uses. Alphas that reflect five years of big gains should not be read as forecasts.
⑤ Positions after the last 13F come from press reports of Burry’s paid newsletter; their size and exact timing can’t be verified. The Section 4 DIVA analysis assumes equal amounts, and post-filing figures are moves in the underlying prices, not option profits or losses.
This piece is for education and information only and does not recommend buying or selling any security or copying any 13F.
Data · 13F: SEC EDGAR 13F-HR, Scion Asset Management LLC (CIK 1649339), Q4 2022 to Q3 2025 · prices: yfinance adjusted close, month-end closes, latest Oct. 9, 2026 · reference ETFs: SPY, IWD, IWF, QQQ · factors: Ken French Data Library, North America FF5 · risk-free rate: US 13-week T-bill, 4.06% annual · Palantir put premium and strike: Burry’s X post as reported by Investing.com (Nov. 12, 2025) · deregistration: Bloomberg, Investing.com (Nov. 2025) · September positions: Stocktwits/Yahoo Finance (Sept. 10, 2026), Benzinga (Sept. 22, 2026), Motley Fool (Oct. 4, 2026) · AI bubble remarks: CNBC (Sept. 28, 2026). Types, factor betas, explained shares, alphas, risk contributions and correlation maps are DIVA Quantizer model outputs.
This article was written with the help of AI.