Michael Burry doubles AI-bubble short: $1.1bn of Nvidia, Palantir, SOXX, QQQ puts and a new Palantir outright short

Michael Burry, the Scion Capital founder whose pre-2008 short of the subprime mortgage market was the basis for the book and film The Big Short, has now doubled his bearish bet against the artificial-intelligence trade and disclosed the expanded positions on his Substack column 'Cassandra Unchained' in a Trading Post entry dated Monday, May 4, 2026 — write-ups by Business Insider on May 5 and TradingKey on May 10 have since aggregated the position-level detail. The new book is concentrated in put options on Nvidia (ticker NVDA) and Palantir (PLTR) with a combined notional exposure of roughly $1.1 billion that represents about 80% of Burry's personal portfolio, together with fresh bearish puts on the iShares Semiconductor ETF (SOXX), the Invesco QQQ Trust (QQQ) and Oracle (ORCL), and a new outright short on Palantir. The position-sizing is the largest single concentration he has carried since the AI short was first established in Q3 2025, and is the first major disclosure since his fund was wound down.

Burry's framing in the Substack post is unusually direct. He writes that listening to financial broadcasts on a long drive made him feel there was "no end to the discussion of AI," with a self-reinforcing narrative increasingly displacing fundamental analysis, and he draws an explicit historical parallel: the recent accelerated ascent of the Philadelphia Semiconductor Index "feels like the final months of the 1999–2000 bubble." On Palantir he goes further than valuation: "I am not just shorting because it is overvalued. I am shorting the business model. I am shorting the entire premise upon which the company rests. I am shorting the CEO." That escalation of the rhetoric — from price-target call to thesis-level short — is the part of the May 4 update that has travelled furthest in the markets coverage.

The specific new positions, line by line

On Nvidia, Burry is now holding put options on 1,000,000 shares struck at $110 and expiring in 2027. With NVDA closing on May 8, 2026 at roughly $215 — a hair under the all-time-high of $217.80 — and the company's market capitalisation now above $5.3 trillion, those puts are deeply out-of-the-money and currently underwater on a mark-to-market basis. Burry's own framing of why he keeps adding to the position: "Nvidia remains one of the cheaper ways to short the AI data-center bubble because of the continued near-unanimous positive view of Wall Street." The newly-purchased puts have strike prices well below the current market price and expire next spring (2027), which signals that he expects a sharp drawdown over the next twelve months, not necessarily a crash inside weeks.

On Palantir, the structure is more elaborate. Burry holds a dual-strike put book — $100-strike puts expiring December 2026 and $50-strike puts expiring June 2027 — and he has now added a direct (non-options) short on top of the puts. Combined exposure on Palantir is the dominant single line in the portfolio. Palantir traded near $161 when Burry first opened the position in Q3 2025; by May 8, 2026 the stock had fallen to roughly $137, a drop of about 34% from its 52-week high of $207. Burry has not yet taken profits despite the gain, and his stated valuation framework for Palantir is brutal: he estimates fair value at "single digits to low double digits", against a market price still north of $130 and a market capitalisation around $350 billion.

Beyond NVDA and PLTR, Burry has expanded his macro-bearish hedges. He bought additional puts on SOXX, the iShares Semiconductor ETF that tracks a basket including AMD, Broadcom, Micron, Intel and Nvidia itself; additional puts on QQQ, the Invesco Nasdaq 100 vehicle whose top weights are Nvidia, Microsoft, Apple, Tesla and Meta; and additional puts on Oracle ahead of an earnings cycle the cloud-infrastructure giant has been telegraphing as AI-revenue-led. Those three add-ons together total roughly 9.5% of his portfolio's short exposure, and the outright short on Tesla that Burry has been carrying for several months sits inside the same bucket alongside the new Palantir outright.

The thesis: 'final months of the 1999–2000 bubble'

Burry's Substack is now titled 'Cassandra Unchained' — a reference to the Greek mythological figure cursed to make accurate prophecies that nobody believes — and the May 4 post is the most assertive single piece he has published since the column launched in late 2025. The two central claims are structural. First, valuation has decoupled from fundamentals: he wrote that Friday's consumer-confidence index hit a record low the same week as the latest non-farm payrolls print, but that "traders remained focused exclusively" on the slightly-better-than-expected jobs number, an asymmetric reaction he reads as evidence that the bull case has stopped responding to negative data.

Second, the parallel to the dot-com terminal phase is exact. The Philadelphia Semiconductor Index, the principal U.S. benchmark for chip stocks, has compounded a gain that on his charts mirrors the steepening curve of the Nasdaq Composite in the November-1999-to-March-2000 window — the period that, in hindsight, is generally identified as the final melt-up before the dot-com crash. Burry's comment in the post — "What a rally, a true cherry on top of a decade-plus explosive run. I am happy being short every single one of those names at current valuations and at this stage of the cycle" — is the cleanest statement he has made of the cycle-stage call underneath his options book.

Scion is wound down — these positions are Burry's personal book

An important context piece for reading the new disclosure correctly: the positions sit in Burry's personal portfolio, not in a regulated fund. Scion Asset Management, his hedge fund, announced its liquidation on October 27, 2025, with capital returned to outside investors by year-end; the SEC terminated Scion's investment-adviser registration on November 10, 2025. The Q3 2025 13F filing — which showed roughly $912.1 million of Palantir puts and $186.6 million of Nvidia puts, together making up about 96.7% of the fund's reported long-equity book — was Scion's final mandatory disclosure. Anything Burry holds today is not subject to 13F reporting, because there is no longer an institutional manager to file one.

That structural change is why Substack has become the primary source for what Burry is doing. His column launched as a paid newsletter in December 2025, and the Trading Post posts — short, structured, position-by-position updates roughly once or twice a week — are now the de-facto disclosure mechanism. The May 4 post is the longest and most explicit of the series to date, and it is also the one that contained the GameStop exit (the first full position he has closed since pivoting from the fund to the newsletter, per his own description in the post).

The hedge portfolio: long beaten-down enterprise software

The mirror image of the AI shorts is a long book in legacy enterprise-software names that the market has, in Burry's reading, unfairly punished as casualties of the AI rotation. The four names he has been adding through April and into May are Adobe (ADBE), Autodesk (ADSK), Salesforce (CRM) and Veeva Systems (VEEV). The shared thesis: each of these companies has its own defensible cash-flow base — creative-tooling subscriptions at Adobe, CAD and AEC software at Autodesk, the world's largest CRM franchise at Salesforce, and life-sciences vertical SaaS at Veeva — and each has been de-rated on a market narrative that AI agents will dis-intermediate the workflows their products sit inside.

Burry's argument is that the price action in those four names reflects panic-selling on narrative rather than deteriorating fundamentals, and that the same Wall Street consensus that is too bullish on Nvidia is now too bearish on the diversified-software cohort. This is the structural symmetry of what the TradingKey write-up labelled his "AI bubble hedge portfolio" — short AI beneficiaries, long AI "victims" — and it is a more complete expression of his thesis than the short book alone, because it requires both ends of the trade to compress for the portfolio to win.

Why GameStop went too: the 'Instant Berkshire' thesis collapsed

The May 4 post also closed out GameStop (GME), which until April had been one of Burry's three largest personal positions alongside JD.com (JD) and Molina Healthcare (MOH). His write-up explains the exit in his own terms: "I sold my entire GME position … the Instant Berkshire thesis was never compatible with greater-than-5x Debt/EBITDA, never okay with interest coverage under 4.0x." The Instant Berkshire idea was Burry's bull case for GameStop CEO Ryan Cohen — that Cohen could deploy GameStop's cash hoard the way Warren Buffett used early-Berkshire float, by acquiring high-quality businesses that compound capital on top of the parent.

The thesis broke when Cohen announced a $56 billion cash-and-stock offer for eBay. Burry's modelled debt math for the combination is the part of the post that did the analytical work: at $56bn, he estimates the combined business would carry a net-debt-to-EBITDA of 5.2x and an interest-coverage ratio of 2.5x; and if eBay pushes back to $65 billion, leverage rises to 7.7x with coverage of just 1.2 to 1.5 times — "a level of debt that borders on distressed." For Burry, the move from Buffett-style compounder to over-levered acquirer invalidated the trade, regardless of how the deal price finally settles.

The 2008 comparison — and the timing risk

The fair counter-question is the timing one, and Burry himself is conspicuous about not dodging it. Before 2008, his CDS-based subprime short sat deep in paper losses for the better part of two years, his outside investors threatened lawsuits, and the position only paid off as the mortgage cycle finally rolled in 2007-2008. The same risk applies now: Nvidia's $5.3 trillion market capitalisation has compounded on a long sequence of earnings beats, the Nasdaq sits at all-time highs, and a single quarter of stronger-than-expected hyperscaler capex guidance would push the AI cohort further away from his strikes before it reverses. The TradingKey write-up notes plainly that the NVDA put position is currently deep in red on a mark-to-market basis.

There is also a structural reason short positions on AI infrastructure can stay underwater longer than the bear case implies. The largest GPU buyers — Microsoft, Alphabet, Amazon, Meta, Oracle and xAI — are all cash-flow-positive incumbents who do not need external financing to keep buying Nvidia silicon; that decouples the chip cycle from the credit cycle in a way that did not characterise 1999-2000. Burry's response, repeated across recent posts, is that valuations are now sufficient on their own — even without a credit accident — to produce a mean-reversion that resolves his options profitably before the 2027 expiries.

What investors actually do with a signal like this

Burry's own framing of the use-case for his disclosures is the appropriate read for any retail investor watching the headlines. In his words, the significance of the "AI bubble hedge portfolio" signal "may not lie in predicting when the bubble will burst, but in reminding the market that when valuations are extremely inflated and narratives are completely detached from fundamentals, the existence of contrarian trading behavior is itself a persistent warning." That is a more defensible framing than a date-the-top call, because it does not require the AI cohort to break on any specific quarterly print.

The dot-com comparison Burry leans on is illustrative of how the path of a melt-up tends to resolve: the Nasdaq Composite climbed roughly 40% in the four months from November 1999 to its March 10, 2000 peak before declining nearly 78% over the following thirty-one months. The asymmetric speed of the down-leg relative to the up-leg is what made the bubble unwind so painful for those who shorted too early and got forced out at the top — and it is exactly the risk Burry is now carrying with NVDA at $215 against an unchanged $110 strike. Whether the AI bet ultimately joins the 2008 mortgage short in the pantheon of correctly-called contrarian trades, or whether it ends up looking like one of his less-fortunate timing calls — Burry has had several of those across two decades — remains, on May 11, 2026, genuinely open.