Michael Burry

Short Thoughts: A Wall Street Titan Hands Us Part V.5 of Our Heretic's Guide to AI's Stars.

The AI capex boom is financed through a fragile, circular web ultimately backstopped by more than $10 trillion of insurance float plus vendor residual-value guarantees on chips. Those guarantees exist because chip depreciation is real (the Nvidia H100 has lost 51% of its value in three years), not because vendors want to offer them. The entire structure rests on a single assumption: sustained AI capital spending. A single season of disappointing AI revenue could lead capital-allocating boards to conclude the highest-conviction bet has shifted, triggering the interconnected guarantees roughly simultaneously and precisely when guarantors are weakest. Burry frames this as the same vendor-financing correlation risk that turned the 2000 telecom fiber glut from a technology story into a systemic one, and as mainstream Wall Street (via Ares) converging on the bearish view of his Heretic's Guide to AI's Stars series.

• US + Bermuda insurance float: a little over $10 trillion
• Nvidia H100 has lost 51% of its value in its first 3 years
• Ares Fall 2026 Alternative Credit Newsletter: pages 2-17 are the key must-read sections
• Ares 'Ares On Hyperscalers And Risk' newsletter PDF: 6.65 MB
• 'Missing $3 Trillion' framing from Heretic's Guide Part IV (Big 5 hyperscalers)
• Historical self-obsolescence analogues cited by Ares: IBM mainframes 1979, Airbus A380 2019, solar PV modules 2014
• Telecom fiber glut of 2000 cited as the vendor-financing correlation-risk analogue
• 3 books on the 1999 bubble were published during 1999

• Insurance float (US + Bermuda, a little over $10 trillion) is the only capital pool deep enough for the AI buildout; Ares notes insurers take on as investment-grade bonds the same catastrophe risk they will not insure as a property liability.
• The Nvidia H100 has lost 51% of its value in its first three years, hard evidence that chip depreciation, not durable pricing stamina, drives the financing structure.
• Residual-value guarantees from Nvidia, Oracle and Google Cloud exist because lenders demand them given depreciation; Burry treats them as a red flag, not reassurance.
• He likens the residual-value and vendor-financing circularity to the 2000 telecom fiber glut, which became a risk-correlation event rather than a technology story.
• The whole web of guarantees, backstops and obligations rests on one assumption, sustained AI capex; a revenue disappointment could trigger guarantees simultaneously when guarantors are weakest.
• Ares's historical self-obsolescence analogues span IBM mainframes (1979), the Airbus A380 (2019) and solar PV modules (2014).
• Part VII will target neo-clouds CoreWeave and Nebius, whose claims of chip-pricing stamina conflict with lender-required residual-value guarantees.
• Burry sees Ares as mainstream Wall Street converging on his bearish view, but warns that wide documentation of a bubble (1999, housing) has never prevented the bust.

[Abridged] Capital Cycle IQ & Forensic Files on the Big 5 Hyperscalers (MSFT, AMZN, ORCL, META, GOOG)

Burry argues the Big 5 hyperscalers are funding a record AI-datacenter capital cycle increasingly through off-balance-sheet leases, purchase commitments, vendor-financing circles and accounting choices that flatter reported economics while pushing risk into the out-years. Market peaks have historically preceded peaks in net investment, and with S&P 500 net-capex versus GDP near a prior-cycle extreme, he places the group late in the capital cycle. Each name shows the same migration toward longer-dated, riskier backlog and rising leverage even as free cash flow turns negative (Amazon) or gross margins compress (Oracle), with useful-life extensions, finance-to-operating-lease reclassification and financing-laden prepayments used to manage optics. Ranking the universe by IV15/IV10 (the prices implying 15% and 10% long-run compounding), he concludes most are too expensive, keeping Alphabet relatively higher in his 'Castle' tier but marking it down for prompt-based LLM search-disruption risk. The setup is a build-up of hidden liabilities and 'bullwhip' demand that he expects to eventually discombobulate both the industry and the stocks.

• MSFT: finance lease liabilities rose $46.2B to $66.6B while bond debt fell $43.2B to $40.3B
• MSFT: $329B of signed-but-uncommenced leases not mentioned on the July 29 call; >1yr backlog +113% vs near-term backlog +37%
• MSFT extends data-center useful life to as high as 25 years (from 15) to clear ASC 842's 75% threshold (13-yr lease implies 17.3-yr minimum)
• AMZN: uncommenced leases + purchase commitments $267B (+81% over nine months); free cash flow roughly negative $8B
• AMZN backlog $200B at 3.8yr WAL (Sep 30 2025) migrated to $496B at 6.4yr (Jun 30 2026); long-term debt raised $746M -> $81.9B YoY
• META: ~$700B uncommenced leases/purchase obligations; VIE max exposure $46B off balance sheet; $28B residual-value-guarantee threshold; $10.8B cash collateral posted
• GOOG: $707B supply-chain purchase obligations, ~$900B total off-balance-sheet exposure (highest of Big 5); long-term debt doubled $46.5B -> $98.2B; $99B equity-linked gains last qtr ($136B over 6 months); Cloud backlog +112% in 6 months
• GOOG/Anthropic: Apollo raised $38B for Google TPUs sold to Anthropic; Alphabet owns 14% of Anthropic and is investing up to $40B
• ORCL: new +$11.4B 'customer prepayments with significant financing component' line; backlog $664B (+$209B YoY), ~$332B beyond FY2029, $106B (~16%) beyond 2031; gross margin down 5 points

• Microsoft's CFO said future data-center leases will 'shift from finance leases to operating leases,' and MSFT extends useful lives to ~25 years to clear ASC 842's 75% threshold, which Burry calls a 'move of convenience,' not a valid accounting basis.
• MSFT left $329B of signed-but-uncommenced leases unmentioned on the July 29 call and analysts did not ask; long-term backlog grew 113% vs 37% near-term, so future revenue increasingly depends on customers who may never raise the capital to pay.
• Amazon's free cash flow is roughly negative $8B while uncommenced leases plus purchase commitments reached $267B (+81% in nine months) and long-term debt issuance jumped from $746M to $81.9B YoY; Burry says 'Amazon is simply too expensive.'
• Meta funds its buildout via VIEs with a $28B residual-value guarantee and $46B of max off-balance-sheet exposure, and posted $10.8B of cash collateral demanded by counterparties 'who know these assets best,' atop ~$700B of uncommenced obligations.
• Alphabet carries ~$900B of off-balance-sheet exposure (most of the Big 5), doubled long-term debt to $98.2B and booked ~$99B of equity-linked gains last quarter; the Apollo/Anthropic/TPU arrangement is a 'seamless circle' of vendor financing.
• Oracle's new +$11.4B 'customer prepayments with significant financing component' line lets it accrete interest into contract liabilities and later recognize it as cloud revenue, which Burry likens to Enron; ~$332B of its $664B backlog sits beyond FY2029 and gross margin fell 5 points.
• Executives lean on 'short-lived assets' and the ability to 'just slow down,' but Burry notes ~85% of a data center's cost repeats each server generation and warns of a 'bullwhip'/barbell-middle that will disrupt the industry and the stocks.
• Burry ranks the CU universe by IV15/IV10 (prices for 15%/10% long-run compounding), keeps Alphabet in the higher 'Castle' tier but marks it down for prompt-based LLM search risk, and previews Part VI on other AI players and leverage schemes.

The Next Big Deadly Fire Nightmare is Being Incubated Right Now by the State of California

Burry, writing as president of his 501(c)(3) nonprofit CRED, argues that California's 'builder's remedy' regime is forcing cities to approve dense luxury housing in Very High Fire Hazard Severity Zones at the top of evacuation routes that already fail. Saratoga's own studies show Pierce Road cannot evacuate current residents before a fire arrives, yet the City Council reversed its Planning Commission's denial and approved the 25-home Masson Estates project out of fear of legal retribution from the State and developer. CRED has sued (CRED v. City of Saratoga) alleging violations of CEQA, the Subdivision Map Act, and the California Constitution's right to safety. Burry frames the 'affordable housing' justification as a fraud — the 'affordable' units are ~$3-4M homes for families earning up to $162,400 — and warns that this pattern is recreating, parcel by parcel, the exact conditions that killed 116 people in Paradise, Palisades and Altadena.

• Camp Fire (2018): 85 deaths, ~19,000 structures burned; fire consumed one football field per second
• Palisades Fire (Jan 7, 2025): grew from ~10 acres to 200 acres in 20 minutes
• Palisades + Eaton fires: 31 deaths, 16,000+ structures destroyed, 100,000 displaced
• Masson Estates: 25-home subdivision in a CAL FIRE Very High Fire Hazard Severity Zone
• City study: residents take 5 hours to evacuate Pierce Road vs 4 hours for fire to arrive
• Fehr & Peers (May 2026) study: project adds ~30 minutes to evacuation, a 13% increase over baseline
• 5 of 25 Masson homes 'affordable' on ~2,100 sq ft lots; 20 market-rate homes up to 7,000+ sq ft on lots up to 2 acres
• San Jose/Sunnyvale/Santa Clara median family income $205,500; 'low income' family of four up to $162,400
• Saratoga existing homes sell ~$1,510/sq ft; a 7,091 sq ft home pencils to ~$10.7M
• Chadwick Heights: 97 units requiring ~800,000 cubic yards of earth moved (90-ft cuts, 70-ft fills) vs the 1,000 cubic-yard municipal limit; 400+ units queued behind Masson on Pierce Road

• This report is an advocacy essay from Burry's nonprofit CRED, not a trading post — it contains no tickers, prices or stated positions.
• CRED filed CRED v. City of Saratoga (Case No. 26CV503751) in Santa Clara County Superior Court on Sept 18, 2026, alleging CEQA, Subdivision Map Act and California Constitution violations.
• Saratoga's own evacuation modeling shows Pierce Road needs 5 hours to evacuate residents while a fire would arrive in 4 — failing even before any new development.
• The City Council reversed its Planning Commission's July 8 denial six weeks later, with members stating on the record they approved out of fear of legal retribution from the State or developer.
• Burry attacks the 'affordable housing' justification as a fraud: 5 of 25 Masson units are 'affordable' but are ~$3-4M homes, defined against a $205,500 area median family income.
• He frames builder's remedy as practiced in Saratoga as 'a mansion-entitlement program sited where the state's own fire maps say no one should add people.'
• The Masson site sits on an active landslide (flagged by Public Works) and adds ~30 min / 13% to evacuation time per the Fehr & Peers study, which outside expert KLD called a life-threatening impact.
• Masson is one of several queued projects on the same corridor: a 255-unit Mountain Winery redevelopment, ~two dozen homes at 12991 Pierce Road, and 97-unit Chadwick Heights requiring 800,000 cubic yards of grading vs a 1,000-yard code limit — 400+ units behind Masson.

Trading Post September 22, 2026

Burry argues the current chip 'shortage' is an artifact of fabs diverting normal-chip capacity to high-bandwidth AI memory, and that as production catches up over the next two years the shortage will blow off and memory will enter another down cycle. With the Nasdaq 100 at historically overvalued, top-heavy levels pushing to new highs, he added to AI/semiconductor shorts (MU, NBIS, SOXX, PLTR) via puts, sized meaningfully, while keeping optionality to cover on further melt-ups. Simultaneously he leaned long into beaten-down consumer/retail names (QXO, BBW, SFM, BIRK, MELI) that have corrected 'tremendously,' taking each to a full position.

• Acer expects Q4 PC price increases of 5% to 20% across products
• DDR4 in oversupply (more sellers than buyers); LPDDR5 / DDR5 9600 still short
• Nasdaq 100 near a new all-time high, described as historically overvalued and top-heavy
• Memory suppliers guiding price increases through 2027; Apple accepted Samsung's contract price increase
• Burry sees the normal-chip shortage blowing off over ~2 years as production catches up
• Long adds (QXO, BBW, SFM, BIRK, MELI) all taken to full positions after large corrections

• Burry's core call is a memory down-cycle: he thinks fabs shifted normal-chip capacity to high-bandwidth AI memory, and that as capacity catches up over ~2 years the shortage 'blows off' - the opposite of the suppliers' guide-to-2027 narrative.
• He leans on Acer CEO Chen Chun-sheng's remarks - Q4 PC prices +5-20%, DDR4 oversupplied, Chinese makers as 'price disruptors' releasing unlimited capacity - as external confirmation, while acknowledging the read is 'fraught.'
• The added shorts (MU, NBIS, SOXX, PLTR) were put on 'in some size' and are expressed through puts he continues to hold, not outright stock.
• These are explicitly tactical shorts: he says he will look to cover if the market takes them to new highs again, signaling defined risk management rather than a permanent bearish anchor.
• He hints at coordinated-signaling among the three big memory suppliers ('under antitrust regulations, companies can only communicate through such signals... those who understand will understand'), and notes even Apple accepted Samsung's price increase.
• On the long side he bought more QXO, BBW, SFM, BIRK and MELI - all names that have 'corrected tremendously' - and took each to a full position, a value-on-weakness stance in consumer/retail.
• The framing sets an overvalued, top-heavy Nasdaq 100 pushing new highs against beaten-down consumer names, a barbell of AI-complex shorts vs. corrected-consumer longs.

Trading Post September 21, 2026: A Number of Value Buys, Including Two New to the CU Universe

Ignoring the AI melt-up ("the woo-hoos"), Burry goes value-hunting and puts on a batch of buys, two of them new to the Cassandra Unchained universe. He initiates Ero Copper as a leveraged play on a structural copper squeeze: large new discoveries have dried up, mines take ~18 years to build, and AI data-center demand is tightening the market on a 2-3 year cycle, so a discounted higher-cost producer with clean accounting and a completing capex program has outsized upside. He also starts QXO, Brad Jacobs's building-products roll-up, preferring its mispriced mandatorily convertible 5.5% preferred (7.4% yield) alongside common, betting on Jacobs's consolidation playbook and a re-rating once housing/rates noise clears. He adds to Temple & Webster, Sprouts and Zoetis. No shorts were covered and no puts sold.

• Ero Copper operations entirely in Brazil; new management (early 2025) delivered on three difficult promises
• Major copper discoveries (>=500,000 tonnes contained) fell from double digits/year in the 1990s-2000s to one or two recently, with none in 2025 (Torsten Slok, Apollo)
• A new copper deposit takes roughly 18 years to reach production -- today's shortfall describes supply in the 2040s
• AI data-center buildout adds copper-intensive load on a 2-3 year cycle vs multi-decade mine lead times
• QXO 5.5% mandatorily convertible preferred now yields 7.4% current cumulative, payable cash or stock
• Bloomberg shows a ~27% preferred-vs-parity gap; Burry says the real gap is only ~4% (floating conversion ratio misread)
• Brad Jacobs (United Rentals, XPO) has run QXO since 2024; stock way down
• Temple & Webster now a fairly large position for him
• No shorts covered and no puts sold this post

• Two brand-new positions this post -- Ero Copper (ERO) and QXO -- both classic Burry value/contrarian setups against an AI-obsessed tape he is deliberately ignoring.
• The copper thesis leans on Torsten Slok's data: near-zero large discoveries (none in 2025) plus ~18-year mine lead times versus AI copper demand on a 2-3 year cycle imply a structural squeeze into the 2040s.
• Burry deliberately chose a higher-cost producer (ERO) over the big miners that already ran, because a higher-cost name has more torque as the squeeze intensifies -- and it still trades at a big discount to peers.
• What made ERO investable to him: new 2025 management delivering on three hard promises, a clean forensic accounting pass, favorable regulation, and a capex program about to start cash flowing.
• On QXO he prefers the mandatorily convertible 5.5% preferred (7.4% current yield) -- a recurring Burry edge: mandatory converts are mispriced with no natural institutional holders and Bloomberg misreads the floating conversion ratio (27% quoted gap vs ~4% real).
• He frames QXO as a Brad Jacobs roll-up bet (United Rentals, XPO pedigree), a stock 'way down' since his 2024 start, expecting a re-rate on execution once housing/rates/diesel noise clears.
• Temple & Webster (TPW AU) is now a fairly large position and he promises a standalone thesis soon -- worth watching as a future deep dive.
• He explicitly did not cover any shorts or sell any puts, so the short book is unchanged; this post is purely accumulation on the long side.

The Heretic's Guide to AI's Stars Part IV: The Big 5 Hyperscalers & the Missing $3 Trillion

The five hyperscalers (Microsoft, Meta, Alphabet, Amazon, Oracle) have quietly loaded roughly $3 trillion of off-balance-sheet obligations — ~$1.2T in mostly non-cancelable uncommenced data-center leases, >$1.5T in purchase commitments, plus SPVs and guarantees — against combined annual earnings of under $400 billion. Under ASC 842 an uncommenced lease creates no asset and no liability until the keys turn, so these commitments sit invisible in footnotes while growing many times faster than the companies themselves. Meanwhile >$400B of capitalized Construction-in-Progress depreciates nothing until switched on, flattering current earnings, and GPUs iterate every 12-18 months while purpose-built halls take 3-5 years to build — a duration mismatch that leaves scant secondary market when demand softens. When the AI capex cycle rolls over (booms historically peak near the midpoint of the spending cycle), these commitments convert into impairments, termination costs and stranded capacity very fast, and the write-offs would dwarf the group's earnings.

• Uncommenced leases across the Big 5 total ~$1.2 trillion, of which at least $857 billion is non-cancelable and off the balance sheet
• Purchase commitments across the Big 5 exceed $1.5 trillion; combined off-balance-sheet contractual commitments ~$2.7 trillion, and above $3 trillion including SPVs, guarantees and backstops
• Big 5 combined earnings over the past year are less than $400 billion
• Microsoft uncommenced leases jumped from $92.7B to $329.1B (10-K); 13-year weighted-average remaining lease term; only 'some' cancelable at 1 year
• Meta uncommenced leases $103.8B prior to $279.0B at June 30, plus ~$68B more signed in July (subsequent event)
• Oracle disclosed $288.0B of uncommenced leases for the first time in its latest 10-Q; Amazon $96.4B->$137.2B; Alphabet $58.5B->$85.2B
• Purchase commitments: Alphabet $811B (the whale), Meta $349B, Microsoft $229B, Amazon $130B, Oracle $34.2B
• Non-depreciating Construction-in-Progress tops $400B: Alphabet $122.8B (29% of gross property), Meta $80.3B (27%), Amazon $71.7B, Oracle $40.0B (+142% y/y), Microsoft estimated $85B-$127B
• Microsoft gross property grew 11x from $38B to $432B over ten years; implied blended asset life 8.25 years on a 6-year server life; MSFT has not disclosed CiP since ~2001
• Microsoft targets 32 GW of data-center capacity by 2032 (~3x current), yet only 2 of its current 12 GW is AI-specific; Oracle carries a $98 price-target spread on a ~$250 stock (39% premium) with vanishingly few sell ratings
• Oracle's reported 121% cloud-infrastructure growth is not apples-to-apples and includes an undisclosed circular-demand share; off-balance-sheet liabilities could reach $4-5 trillion by 2028

• Uncommenced leases are the tell: ASC 842 keeps a non-cancelable 20-year data-center lease off the balance sheet until it commences, so ~$1.2T of Big 5 lease obligations (>=$857B non-cancelable) sit only in footnotes.
• Combined off-balance-sheet commitments (~$2.7T of leases + purchases, >$3T with SPVs/guarantees) dwarf the group's <$400B annual earnings and are growing many times faster than the companies.
• Alphabet is the purchase-commitment whale at $811B — far above Meta ($349B), Microsoft ($229B), Amazon ($130B) and Oracle ($34.2B) — an unusual physical-supply-chain risk for a former AdTech name.
• Microsoft's uncommenced leases tripled to $329.1B and it hasn't disclosed Construction-in-Progress since ~2001; Burry back-solves ~$85B-$127B of hidden CiP from its depreciation gap.
• Over $400B of capitalized CiP across the group depreciates nothing until switched on, flattering current earnings while Nvidia GPUs 'economically disappear' in warehouses.
• Duration mismatch is the fault line: GPUs iterate in 12-18 months but custom-cooled halls take 3-5 years to build, and a purpose-built data center has no deep resale market when unneeded (unlike Meta's subleasable offices).
• Oracle is the wobbly one — 121% cloud-infra growth is not apples-to-apples and carries undisclosed circular demand; Burry likens its inscrutable dual-purpose balance sheet to pre-GFC AIG, and flags its $98 target spread with almost no sell ratings as a contrarian top signal.
• Booms peak near the midpoint of capex cycles; when the AI buildout rolls over, off-balance-sheet commitments convert to impairments, termination costs and stranded capacity fast, with absolute write-offs potentially exceeding anything else on the tape.

The Sickle Cell Development Fund

This is Cassandra Unchained's September 2026 Charity of the Month post, not an investment note. Burry draws on his time as a medical intern in Memphis to spotlight sickle cell disease, argues that although cures exist (bone marrow transplant, 2023 gene therapies) the real problem is now access given costs of $2.2M-$3.1M per patient, and champions Dr. Michael DeBaun's Vanderbilt-Meharry program that uses cheap transcranial Doppler screening and generic medicine to prevent strokes in at-risk Nigerian children at a fraction of the cost. He commits 5% of the month's gross revenues to the fund.

• Approved gene therapies for sickle cell list at $2.2M-$3.1M per patient
• Bone marrow transplantation, curative for decades, costs hundreds of thousands of dollars
• ~150,000 children born with sickle cell disease annually in Nigeria
• Over 11% of children with sickle cell suffer a stroke before age 20 absent intervention
• Cassandra Unchained will donate 5% of September gross revenues to the fund

• Not an investment post — this is Cassandra Unchained's September 2026 Charity of the Month, with no stated positions, prices or tickers.
• Burry reframes sickle cell as an access problem, not a science problem: cures exist but gene therapies list at $2.2M-$3.1M per patient and are confined to a few Western centers.
• The featured intervention is high-leverage philanthropy: transcranial Doppler screening plus generic medicine to prevent strokes in Nigerian children at a fraction of transplant/gene-therapy cost.
• Scale of need is concentrated in sub-Saharan Africa — roughly 150,000 children born with sickle cell annually in Nigeria, where over 11% would stroke before age 20 without intervention.
• Cassandra Unchained is donating 5% of September gross revenues to Dr. DeBaun's Vanderbilt-Meharry Sickle Cell Development Fund.

Shorting the Dollar by Drinking Free First Growth Bordeaux For Life

Burry argues bonded fine wine held in Europe (e.g. at London City Bond) is a scarce, borderless real asset that functions structurally as a long-horizon short on the U.S. dollar. Fine wine has fallen ~25-30% from its October 2022 peak, offering a cyclically depressed entry; academic data show ~4.1% net real returns over 113 years, and buying well below market plus continuous supply destruction (bottles are consumed) adds to that. Because the wine is priced by a multi-currency global bid and held outside U.S. jurisdiction, a 30-40% dollar decline over two decades layers a large FX kicker on top of the wine's own return, while European-vs-U.S. price spreads (45-80%) let a buyer effectively drink First Growth Bordeaux 'for free' by capturing the arbitrage through consumption. He frames it as an uncorrelated hedge against fiscal debasement and AI/quantum-era risks to digital assets, and discloses he has executed and continues to execute the strategy.

• Fine wine down ~25-30% on the Liv-ex indices from the October 2022 peak (deepest broad correction of the modern era)
• The Price of Wine (Dimson, Rousseau, Spaenjers, JFE Nov 2015): 5.3% annual real return before costs, 4.1% net of storage/insurance, five First Growths 1900-2012 (36,271 prices)
• 1900-2012 British real returns: equities 5.2%, stamps 2.8%, art 2.4%, govt bonds 1.5%, T-bills 0.9% — fine wine beat all collectibles/fixed income
• DXY fell 10.8% in H1 2025 (worst half since 1973), incl. -9.6% vs Swiss franc in one quarter
• Dollar share of allocated FX reserves fell from 72% (2001) to just under 57% (Q3 FY2025) — first sub-57% reading since 1995
• Wine vs dollar: ~-0.74 correlation across overlapping 5-year periods (block-bootstrap p=0.01)
• Liv-ex 100 vs DXY: essentially zero monthly correlation; wine returned avg 10.7% in 9 down-dollar years vs 4.4% in 13 up-dollar years
• Same bottle runs 45-80% more in America than at the European trade level (Margaux 2015 ~$1,999 US vs ~£1,120 UK); 15% EU wine tariff since 2025
• US Treasury flagged ~$867B in FY2026 interest through May 2026; fiscal policy called long-term unsustainable
• Target buy discounts of ~18-20% to European trade level; bought ~40 of ~700 wines considered this year
• Lafite produces up to ~20,000 cases/year; Bordeaux fell from ~95% to ~38% of the traded market
• Liv-ex Fine Wine 100 vs S&P 500: +0.136 monthly correlation (p=0.02) — near-perfect diversifier that still roughly matches S&P performance

• Burry pitches bonded European fine wine as a structural, unleveraged, no-decay short on the dollar — a real asset priced by a multi-currency global bid and held outside U.S. jurisdiction.
• Entry timing is the edge: wine is ~25-30% below its Oct-2022 peak, the deepest broad drawdown of the indexed era, so today is a cyclically depressed (non-random) entry point he thinks can lift long-run real returns toward 6-8%.
• The 45-80% Europe-vs-US price gap lets a holder capture 'free wine for life' through consumption/gifting (carve out 15-20% of the book), while the day-one discount covers the entire drinking tranche.
• Currency kicker: a 35% dollar decline over 20 years turns a flat wine market into ~+54% in dollar terms; he shows ~-0.74 wine/dollar correlation over overlapping 5-year windows (p=0.01).
• His 2010 Lafite 'autopsy' warns that First Growth wineries now price in future appreciation at release and dump ex-château supply at industrial scale (Lafite up to ~20,000 cases/yr) — so buy distressed later, never at release.
• Provenance is everything: buy from large reputable dealers and store at a third-party bonded warehouse (London City Bond) to defeat the three frauds (wine that was never yours, wine that never existed, wine that isn't what the label says).
• He buys selectively — ~40 of ~700 wines considered this year, targeting ~18-20% discounts to the European trade level — and treats the 'fireflies' shelf (e.g. Pavie 2021) as the only investable one.
• Framed as macro insurance: dollar reserve share down from 72% to <57%, DXY -10.8% in H1 2025, ~$867B FY2026 interest — with wine a near-perfect S&P diversifier (+0.136 correlation) that still roughly matches equity returns.

Trading Post Friday September 11, 2026: Selling Out of One to Pile Into Another

This is a relative-value rotation: Burry sold his entire Flutter (FLTR) position because it looks less attractive at current prices than other stocks that have fallen further, and redeployed the majority of the proceeds into lululemon (LULU) under $100. He argues LULU is highly profitable with a badly compressed valuation whose troubles stem from poor management, making it an easy target for an acquirer who could run it better — a take-private he views as plausible enough to act on with urgency. He also added to Zoetis (ZTS) in the $72s as a cheap, basing name.

• Sold Flutter (FLTR) in full
• Added to lululemon (LULU) under $100
• Bought more Zoetis (ZTS) in the $72s
• LULU deep dive 'An Ode to Bad Management' published ~3 months prior (420 likes, 163 comments)
• Next Heretic's Guide (Part 4) due early the following week, after Oracle's results closed out earnings

• Burry exited Flutter (FLTR) entirely — not on a soured thesis but because other stocks have fallen to more attractive valuations, a relative-value rotation.
• The bulk of the Flutter proceeds went into lululemon (LULU) under $100, concentrating the book into a single high-conviction idea.
• His LULU case is a bad-management-plus-takeout thesis: profitable, compressed valuation, troubles caused by management, so an acquirer could buy and run it better; he flags possible take-private with 'urgency' but stresses no inside information.
• He added to Zoetis (ZTS) in the $72s as 'very cheap,' noting it is basing on high volume but may bleed lower first — buying value without calling the exact low.
• The move shows Burry running a concentrated book, willing to sell a full position to fund a higher-conviction one rather than trimming at the margin.
• Earnings season wrapped with Oracle's print; he signaled the next Heretic's Guide (Part 4) for early the following week.

Shorting the Dollar by Drinking Free First Growth Bordeaux

Burry argues that European bonded fine wine, bought during its deepest drawdown in the indexed era, is a low-decay, unleveraged short on the US dollar and a hedge against fiscal debasement and AI/quantum-driven digital-asset risk. Because top wines are priced by a global multi-currency bid and stored in bond outside US jurisdiction, a weakening dollar translates directly into dollar-denominated gains on top of the asset's ~4% real return and its built-in supply depletion. Buying 100-point First Growths and top Super Tuscans at 18-20% discounts in a cyclical trough stacks a valuation discount, a currency kicker and an aging 'dividend'. The consumption arm is effectively free: the day-one discount to market covers a lifetime of gifting and special-occasion bottles, while the bulk stays in bonded storage as the investment. The downside case — dollar strength — still leaves scarce real assets compounding at mid-single-digit real rates with near-zero correlation to equities.

• Fine wine ~25-30% below the October 2022 Liv-ex peak; deepest broad correction of the modern indexed era, and now below long-term trend
• The Price of Wine (Dimson, Rousseau, Spaenjers; JFE 2015): 36,271 auction/dealer prices for the five First Growths, 1900-2012; 5.3% annual real return before costs, 4.1% net of storage and insurance
• 1900-2012 UK real returns: equities 5.2%, stamps 2.8%, art 2.4%, gilts 1.5%, T-bills 0.9%; wine 113-yr nominal mean 10.9% vs equities 11.2%
• Depressed-trough entry could lift long-run real returns to 6-8%; a 35% dollar decline over 20 years adds a ~+54% dollar kicker on a flat-to-4% real wine book
• Dollar share of allocated FX reserves fell from 72% (2001) to <57% (Q3 FY2025), the first sub-57% reading since 1995; gold is now the second-largest reserve asset
• DXY fell 10.8% in H1 2025, its worst half since 1973, including 9.6% against the Swiss franc in one quarter
• Over overlapping 5-year windows, Liv-ex Fine Wine 100 shows -0.74 correlation to the dollar index (p=0.01); in 9 DXY-down years wine returned 10.7% avg vs 4.4% in 13 DXY-up years
• Same bottle runs 45-80% more in the US than at the European trade level (Margaux 2015 ~$1,999 US vs ~$1,120 UK trade); 15% EU-wine tariff in force since 2025
• Burry targets 18-20% purchase discounts to the European trade bid; of ~700 wines considered this year he bought ~40 (buys roughly 1 of 20 'fireflies')
• US Treasury flagged fiscal policy as long-term unsustainable; ~$867B of FY2026 interest through May 2026; Bordeaux has fallen from ~95% to 38% of the traded fine-wine market

• Burry is not trading a stock here — the 'position' is physical bonded fine wine in Europe, which he frames as an unleveraged, no-margin, no-decay short on the dollar layered on top of a ~4% real return and continuous supply depletion.
• The tell for value is the European trade bid, never US retail: he rejected Latour 2000 pitched at '35% off' because it was discounted to the inflated American price, which he calls 'no information at all'.
• The 'free wine' works only because the day-one discount (he targets 18-20%, on a book carrying a ~22% value bonus) covers the entire cost of the bottles pulled for drinking and gifting; the rest stays in bond as the investment.
• His dollar-hedge evidence firms up only over long horizons — no monthly correlation, but -0.74 over overlapping 5-year windows (p=0.01) and 10.7% wine returns in dollar-down years vs 4.4% in up years.
• The 2010 Lafite 'autopsy': a perfect-score First Growth returned less than nothing over 15 years because, post-GFC, the château now prices at release to capture the appreciation itself — so gains accrue to distressed later buyers, not original subscribers.
• Supply/rotation matters: Lafite makes up to 20,000 cases a year (vs 6,000-8,000 across all seven Burgundy grand crus) and periodically releases ex-château stock, and Bordeaux has fallen from ~95% to 38% of the traded market — the rotation away is the opportunity.
• Fraud defense is structural: buy from large reputable dealers (AVU, Justerini & Brooks, Bordeaux Index, Berry Bros. & Rudd, etc.) and force third-party bonded storage at London City Bond with case photos and rotation numbers to defeat the three fraud types.
• Macro glue: he ties tariffs to reserve status — a weakening dollar makes the 15% EU-wine tariff politically unsustainable, so the American bonded owner effectively holds a call option on future tariff reversal while capturing the currency move regardless.

Trading Post September 9, 2026: Another Quick One

Burry is pulling in risk heading into the fall. He closed his December 2026 NVDA and Palantir puts outright to sidestep the accelerating time decay that hits short-dated options in September, and rather than rolling them he let the exposure come off. He then trimmed every remaining long and short across the book to raise cash, keeping the overall sizing order intact rather than making directional bets. He retains longer-dated 2027 puts on Palantir and QQQ as standing downside hedges. The message is defensive positioning and dry powder ahead of what he expects to be an interesting, volatile fall market — with a falling dollar index flagged as the subject of an upcoming piece.

• Sold 100% of December 2026 NVDA puts (no roll)
• Sold 100% of December 2026 PLTR puts (no roll)
• Trimmed every long and every short position in the portfolio
• Largest longs (order): LULU > MOH > MELI > TPW AU > ZTS > SFM > JD > BIRK > ADBE > HCA > FISV > FLUT > BBW > VEEV > FNMA/FMCC > PYPL
• Short book (order): ORCL > PLTR > NBIS > NVDA > SOXX > MU > CAT > CRWV
• Still holds 2027 puts on PLTR and QQQ (left untrimmed)
• Net result: raised cash, happy to sit on it
• Watching the US dollar spot index (DXY) fall again

• Burry closed his December 2026 NVDA and PLTR puts entirely — the move was about avoiding September theta decay on short-dated options, not a change in his bearish view on either name.
• He deliberately did not roll the closed puts, signaling he wants less gross exposure overall rather than a rotation.
• Every long and every short was trimmed, but the relative sizing order was left unchanged — this is de-grossing, not repositioning.
• His largest long remains Lululemon (LULU); his largest short remains Oracle (ORCL).
• The AI-infrastructure short book (ORCL, PLTR, NBIS, NVDA, SOXX, MU, CRWV) is intact but smaller after the trims.
• He keeps 2027 puts on Palantir and QQQ untrimmed, so his index and PLTR downside hedges now sit in longer-dated options.
• The trims raised cash he is 'happy to sit on,' explicitly positioning defensively for a volatile fall.
• A falling US dollar spot index (DXY) is flagged as the subject of an imminent post — a macro thread to watch next.

Trading Post 9/4/2026: lululemon (LULU) Is a Fat Pitch Under $100

lululemon's fundamentals have deteriorated sharply — a -9% global comp, leggings down ~20%, and mid-single-digit operating-margin guidance, with the headline EPS 'beat' flattered by a one-time tariff benefit. But Burry argues the guidance has been deliberately kitchen-sinked ahead of an incoming CEO who inherits the lowest possible bar, while the stock at ~$99 trades just over 2x tangible book with $1.3B cash and no debt for a brand he still believes in. Under every scenario in which LULU ever grows again, $99.50 sits below his IV15 and implies 15-20% annual returns over 15-20 years, making it a 'fat pitch' under $100. He is adding LULU and flags JD (owned) and Alibaba (to be bought) as the only other clear 'super fat pitches' in his universe.

• LULU global comp -9%; leggings sales down ~20%
• Operating margin guided to mid single digits
• Negative growth in both China and the US despite store expansion
• Company retired 7% of shares at $142, below Burry's IV8-IV10 intrinsic range
• $1.3B cash and no financial debt
• Shares at ~$99 = just over 2x tangible book value
• $99.50 below his IV15; est. 15-20% annual returns over 15-20 years
• Buyback/entry: plans to add LULU under $100

• Burry's IV15 model on LULU fell sharply after the latest 10-Q/call/8-K: -9% global comp, leggings -20%, and mid-single-digit operating-margin guidance.
• He argues the headline EPS beat was really a miss once a one-time tariff benefit is excluded.
• The bull case is a 'kitchen sink' setup: he thinks guidance was deliberately loaded with bad news so the incoming CEO starts from the lowest possible bar.
• Tells he cites: front-loaded marketing expense, guidance excluding re-orders of products that are selling, and suspicious 'expense timing' language.
• Valuation: ~$99 is just over 2x tangible book, with $1.3B cash and no financial debt; the company bought back 7% of shares at $142, above where it trades now and below his IV8-IV10 range.
• Under scenarios where LULU grows again, $99.50 is below his IV15 and implies 15-20% annual returns over 15-20 years — he calls it a 'fat pitch' under $100 and plans to add.
• He compares the wager to betting ANF, Ralph Lauren, or LULU (in 2017) would never grow again — bets that proved wrong.
• Beyond LULU, the only other 'clear super fat pitches' he names are JD (owned) and Alibaba (plans to buy soon).

Short Thoughts - Cassandra in Flames?

LULU, Burry's largest position, has fallen roughly 20% after a weak quarter he expected, and he treats the drop as a buying opportunity rather than a reason to exit. His logic: the stock is now well under his IV15 intrinsic value, the thesis is intact, and a new CEO arrives this month, with a private-equity takeout as a backstop. He frames enduring pain in a top holding as his M.O., citing Avanti in 2001 and recent Cassandra Unchained names (Adobe, Molina, Veeva) that crashed then rebounded sharply, versus Tailored Brands as a rare permanent loss. His decision rule at a ~20% drawdown is binary: buy more or sell, and here he chooses to buy aggressively under $100.

• LULU is Burry's largest position
• LULU down ~20% from his last purchase; trading under $100 after hours
• Stated plan: buy aggressively if LULU trades under $100
• LULU trades well under IV15 (his intrinsic-value measure)
• New LULU CEO arrives September 2026
• Avanti (2001): bought at $12, fell to the $2s, acquired at $22/share
• Adobe, Molina and Veeva cited as prior CU names that crashed then rebounded
• Tailored Brands cited as a rare permanent loss of capital via bankruptcy

Trading Post August 28, 2026

After a near-doubling in Veeva and a collapsed deal in PayPal, Burry books his first real sells of the week. He trims Veeva to a ~3% 'house money' stake because its valuation has stretched past his IV15 fair value, harvesting capital to redeploy into more spring-loaded ideas, while keeping a small position since it is still cheap on IV8. He holds a reduced ~3% PayPal position after the stock fell into the $53s, planning to add in the mid-$40s. The through-line is valuation discipline and after-tax return comparison: he sells only when a name is uncomfortably expensive and better opportunities exist, not to chase long-term capital-gains treatment.

• Veeva (VEEV) nearly doubled over the last few months
• VEEV trimmed to a ~3% position ('house's money')
• PYPL fell nearly $8 to the $53s after a deal collapsed
• PYPL held at a ~3% position; would add in the mid-$40s
• Taxes take 20-50% of the gain on a sale, factored into sell decisions
• Cassandra Unchained software & payments universe now 50 stocks (the SW50)
• Uses IV15 as the primary valuation guide, plus IV8 for a nearer-term read
• SW50 recently added Synopsys, Cadence, JFrog and Samsara

• Burry's first real sells of the week are trims, not exits: both Veeva and PayPal are cut to ~3% 'house money' stakes rather than closed.
• Veeva was sold down after nearly doubling and running past his IV15 fair value, but he keeps a stake because it is still cheap on IV8.
• He explicitly warns against holding an uncomfortable position just to reach long-term capital-gains treatment — 'a good way to get burned.'
• Taxes (20-50% of the gain) are treated as a first-class input to sell decisions, compared against the after-tax return of alternative ideas.
• PayPal fell nearly $8 into the $53s after a deal collapsed; he had pre-emptively reduced it and now holds ~3%, with a stated add point in the mid-$40s.
• Capital harvested from Veeva is earmarked for 'more spring-loaded positions' — signaling he sees better risk/reward elsewhere.
• The Cassandra software & payments universe (SW50) now spans 50 names after adding Synopsys, Cadence, JFrog and Samsara.
• No new longs or shorts this post — it is a discipline/housekeeping note showcasing his valuation framework in action.

Trading Post August 27, 2026

Burry treats this as a real-time Trading Post. On Nvidia he monetizes a very short-term call hedge that worked (up 50-60% into earnings), legging out because the event has passed while his long-term bearish thesis on the stock is intact. On Build-a-Bear, a ~28% earnings-day drop trips his 20%-down discipline; rather than react, he holds his existing long, points to the stock trading below historic IV15 as valuation support, and defers any add until he can read the 10-Q — arguing the brand-dilution/wholesale problems that caused the guide-down are fixable under new management.

• Sold half of NVDA calls up 50-60%; position was established <24h earlier as an earnings hedge
• Expects to sell the remaining half of the NVDA calls (no good reason to hold; remains long-term bearish)
• BBW down ~11 points (~28%) intraday, triggering his 20%-down rule
• BBW commercial/wholesale segment guide cut to flat from +20% previously
• BBW international franchising down 33%; partner operators also declined
• Walmart wholesale relationship non-renewed
• New BBW CEO as of May 2026; Chief Growth Officer (behind the Squishmallows-chasing 'plushy push') fired without cause
• BBW growth plan intact: 50+ new locations, capex target ~$25M (a micro cap)
• No impairment charges or inventory write-downs in the quarter; 10-Q not yet filed (expected today/tomorrow)
• BBW trades below historic IV15

• The NVDA calls were an explicit, sub-24-hour earnings hedge — not a directional bull bet; Burry booked +50-60%, sold half, and plans to exit the rest, keeping his long-term bearish stance on the stock.
• He openly concedes 'little standing in the way of further stock appreciation' in NVDA and that the print did no thesis damage ('no head shots'), a notably candid acknowledgment that the bull narrative can persist.
• BBW's ~28% drop tripped his 20%-down rule, but he stresses the rule mandates a decision, not immediate action — he is holding, not panic-selling.
• The bear case that caused the guide-down is, in his read, self-inflicted brand dilution: chasing Squishmallows with generic pre-made plush undercut the 'build-a-bear' experience for a temporary sales pop.
• Management change is central to his bullish setup: new CEO since May and the Chief Growth Officer behind the plush push fired without cause signal a strategy reset.
• He flags a tension — the growth plan (50+ new stores, ~$25M capex) stays intact even as commercial revenue is guided to flat — as something the 10-Q needs to justify.
• Valuation anchor: BBW now trades below historic IV15 (his intrinsic-value screen), which is why he is inclined to add rather than exit — pending the 10-Q.
• He explicitly separates a clean-looking earnings release (no impairments or write-downs) from the guidance shock, resisting the 'kitchen sink' interpretation until the full filing confirms it.

Trading Post August 26, 2026 - Nvidia, Friends, and the Rebel Alliance

Burry believes Nvidia's price is incongruent with the market's own narrative: on the surface it looks wildly undervalued (low PE, monopoly rents), yet the stock is only treading water, which he reads as the market quietly disbelieving the story. His analysis points to a short runway for Nvidia's monopoly power, falling margins, and a capex/investment focus that will starve shareholder distributions — so he is fundamentally short NVDA and the broader AI-infrastructure complex (ORCL, PLTR, NBIS), plus cyclical CAT. Because that short-and-put position is so large, he bought December NVDA calls as a pure earnings-day hedge, not a directional long. On the other side of the book, his 'Rebel Alliance' of longs — Birkenstock and the GSEs Freddie Mac and Fannie Mae — got added to.

• NVDA short-plus-put position is 3.5-4% of portfolios
• Bought December NVDA calls, strike mid-to-high $200s, low implied vol, premium paid 'single digits'
• Total short stock position over 21% of portfolios, not including puts
• Added BIRK in the mid-$30s to a full position
• Added FMCC in the mid-$5s; large GSE position with FNMA
• Topped off shorts this week in ORCL, PLTR, NBIS, NVDA and CAT
• References NVDA revenue chart with a 100% year-over-year growth line
• Puts described as far out-of-the-money, distant expiry, minimal current market value

• The December NVDA calls are explicitly a hedge on an earnings-day coin flip, fully offset by his short-plus-put position of 3.5-4% of portfolios — not a change of heart on Nvidia.
• His core NVDA bear case: theoretical (DCF) value well below market, a short runway for monopoly power, falling margins, and capex crowding out shareholder distributions.
• He reads NVDA merely 'treading water' despite a low PE as evidence the market itself does not believe the bull narrative.
• The short book is large and broadening: total short stock position now exceeds 21% of portfolios, excluding puts.
• This week he added to shorts in ORCL, PLTR, NBIS, NVDA and CAT — the AI-infrastructure/hyperscaler-capex complex plus a cyclical.
• On the long side, the 'Rebel Alliance' grew: BIRK added in the mid-$30s to a full position, FMCC added in the mid-$5s as part of a large GSE position with FNMA.
• His puts are deliberately far out-of-the-money and distant in expiry — asymmetric, leveraged and directional with minimal current mark.
• He admits a mediocre track record trading single-day earnings options, framing this NVDA call purely as insurance around a large short.

Trading Post & Short Thoughts to the Max, August 18, 19, 20, 2026

Burry argues US GDP is dangerously dependent on a single, debt-financed AI data-center buildout even as the economics that justify it are eroding. Citing new academic work on 'intelligence per watt,' he says small local/open-weight inference models now handle the vast majority of queries far more cheaply, so billable inference will migrate off hyperscaler GPUs and expose a duration mismatch in six-year server depreciation. That undercuts the returns underwriting the buildout, which he likens to pre-GFC leverage, fraud and off-balance-sheet financing (now via captive insurers). He positions accordingly: shorting AI-infrastructure and richly valued newer names (CRWV, MU, SOXX, PLTR) plus far-OTM index/NVDA/PLTR puts, while building large longs in cheap, misunderstood consumer and healthcare businesses (LULU, MELI, Temple & Webster, SFM, BIRK, BBW, MOH).

• BIRK increased to a 5.2% position as the stock fell into the $35s
• BBW new full long in the mid-to-upper $30s
• TPLWF bought in the low $3s; Temple & Webster now a ~9% position
• SFM to a full 5.2% position in the high $70s
• LULU added in the $115s to a very large 17.4% position; earnings Sept 3
• MOH added at $201, now a 9% position
• PYPL trimmed to a 4% position (trading above deal price); MELI added in the $1890s to 12%
• Total shorts excluding puts a little over 21% of the portfolio; puts ~6.5% of the portfolio
• Local-model coverage rose from 23.2% (2023) to 71.3% (2025); 88.7% of single-turn queries handled locally
• Intelligence-per-watt improved 5.3x in two years (3.1x model, 1.7x hardware); private non-residential construction ex-data-centers fell 7.9% YoY in June

• Burry's central macro claim: with private non-residential construction ex-data-centers down 7.9% YoY in June, US GDP growth is riding almost entirely on the AI data-center buildout — a single-factor dependence he compares to pre-GFC housing.
• He flags captive offshore/onshore insurers absorbing risky or fraudulent paper and off-balance-sheet SPV financing as the hidden leverage in the AI buildout, echoing the BIS 2026 annual report.
• The 'Intelligence per Watt' paper is his evidence base: 88.7% of single-turn queries handled by small local models, IPW up 5.3x in two years, local coverage up from 23.2% (2023) to 71.3% (2025).
• His 'duration mismatch' point: six-year server depreciation is too long because inference migrates to local chips, breaking the billable-inference-per-GPU math behind debt-financed hyperscaler capex.
• Shorts are concentrated in newer, richly valued AI-infra names (CoreWeave new short, added Palantir) plus long-public semis (Micron, SOXX); total shorts ex-puts now just over 21% of the portfolio.
• The put book (~6.5% of the portfolio) is far-OTM NVDA, PLTR and QQQ, mostly 2027 expiries with a portion of PLTR puts expiring December 2026.
• Long book is a cheap-consumer/healthcare barbell: LULU (17.4%), MELI (12%), Temple & Webster (~9%), MOH (9%), SFM (5.2%), BIRK (5.2%), plus a new BBW full position.
• He is running mid-high-teens percent margin to accommodate a larger long book against the big short, but says he'll bring margin back toward 0–10% and warns retail investors away from shorting and margin.

Acres of Diamonds

This is Cassandra Unchained's August 2026 Charity of the Month, not an investment post. Burry vets Acres of Diamonds, a small Duvall, Washington charity that houses homeless mothers and children, by pulling its Form 990 and applying the same forensic-disclosure lens he uses on companies. He concludes the filings are clean and non-disqualifying, program efficiency (79%) clears every major watchdog threshold, and donor support is genuine at 94.5% of revenue. As promised, 5% of Cassandra Unchained's August gross revenue will be donated to the charity.

• 44 families and 50 children housed last year
• Public support = 94.5% of revenue
• Fundraising cost ~$10.80 per $100 raised
• Executive director base salary $54,000
• ~$3.3 million total spent by the charity
• ~$50,000 spent per family
• Contributions doubled over 5 years to $4 million in 2025
• Expense ratio 21% (79% program efficiency)
• Clears CharityWatch 75%, Charity Navigator 70%, BBB 60% standards
• In operation for three decades

• Charity-of-the-month post, not an investment thesis — no tickers or positions.
• Burry applies forensic-filing scrutiny to charities, reading the Form 990 Part IX just as he would a 10-K.
• Acres of Diamonds: 94.5% public support, ~$10.80 fundraising cost per $100 raised, $54k ED salary.
• 79% program efficiency clears CharityWatch (75%), Charity Navigator (70%) and BBB (60%) thresholds.
• Contributions doubled in five years to $4 million in 2025; ~$50,000 spent per family served.
• He deliberately seeks small, effective charities without celebrity sponsorship where exposure moves the needle.
• 5% of Cassandra Unchained's August 2026 gross revenue will be donated to the charity.

Trading Post August 13, 2026: Welcome to the Sausage Factories

Burry argues the AI trade is being financed circularly — AI companies funding each other to buy from each other, so injected capital returns as revenue and Wall Street rewards the resulting growth — and that the capital funding it is increasingly debt, which puts the bubble on a clock with a concrete cost of capital. His base case for the turn is 2028, when compute becomes too much, though he notes markets historically catch on well before that. In the meantime the rally has tipped his short book from roughly breakeven into a loss, so he is de-grossing: selling volatility where it still exists, buying cheap volatility where he can, and covering the shorts that have already paid. The signature trade is rotating out of expensive-IV SOXX puts into a larger, longer-dated QQQ put position at 25 IV, betting that compressed index volatility spreads on further rallies. He simultaneously trims every long to raise ~12% cash, explicitly preparing for a larger market fall while keeping his short bias intact.

• JD revenue -2.9%, the first decline since its 2014 IPO; stock down almost 10% on the day
• QQQ put roll: January 2027 strikes in the mid-high $500s rolled to June 2027 strikes in the mid-high $600s, at a substantial debit
• Volatility trade moved from 45 IV in the SOX to 25 IV in the QQQs
• QQQ put position now about 6% of the portfolio
• SOXX short is the largest short at about 7% of the portfolio
• Micron short increased on its move back up near $1,000
• Caterpillar short trimmed by 25%
• Cash position raised to roughly 12% via an even sell-down across all longs
• Top four longs (ZTS, MELI, JD, ADBE) all about the same size, separated by basis points; full positions range 5% to 8%
• Base case for the AI bubble turning: 2028, when he thinks compute becomes too much

• Burry puts a date on it: his base case for the AI bubble's turn is 2028, when compute becomes too much — but he notes markets historically catch on well before the actual turn.
• The reason debt matters more than the circularity itself is that it gives the bubble a concrete cost of capital, converting a sentiment story into a clock.
• The core trade is a volatility rotation, not a directional change: out of 45 IV SOXX puts and into 25 IV QQQ puts, on the view that compressed index vol spreads once rallies force it wider.
• He extended duration as well as strike — January 2027 mid-high $500 strikes rolled to June 2027 mid-high $600s at a substantial debit, buying time and moneyness rather than reducing the bet.
• The short book is now ranked SOXX (~7%, largest) > Micron > Nebius > Nvidia > Oracle > Palantir, with the QQQ puts a separate ~6% sleeve.
• Micron was the one counter-trend add, increased on its move back near $1,000 while nearly everything else was cut.
• Oracle is shorted in shares rather than options purely because its puts remain too expensive — a mechanical, not thematic, distinction.
• He treats JD's first-ever revenue decline (-2.9%) as an artifact of lapping a subsidy bonanza, not deterioration, and holds it as a top-four long on the view that China's Great Depression has already hit its nadir.
• The ~12% cash position was raised by shaving every long a little rather than exiting any — preserving the full long roster while explicitly preparing for a larger fall.

Trading Post August 12, 2026: More Buys, More Shorts

Nebius is inadvertently publishing the evidence against its own asset base: wholesale compute is sold in power terms because power is scarce, yet short-dated compute clears at roughly twice the price of multi-year compute. That is backwardation, and backwardation exists because the underlying commodity is depreciating. Power does not depreciate, so either the GPU is depreciating, the customer is depreciating, or — Burry's view — both. Management's decision to deliberately withhold 2027 capacity to keep selling short-term contracts confirms it knows the value sits at the front of the curve, even as it lengthens server depreciation from four years to five and books the difference. Burry therefore presses the AI-complex short book (ORCL, MU, NBIS, SOXX puts) into strength while adding to the two non-AI longs, MELI and ZTS.

• Added to MELI long at $1,850
• Added to ZTS long at $73.60
• Sold short more ORCL at $152 and more MU at $924
• Sold short more NBIS at $247
• Nebius closed 4 landmark deals in the quarter — Reflection, Cohere, a scaled US lab and a large US quant trading firm — averaging more than $1 billion each
• Nebius midterm (1-3 year) contracts yield $20-25 million per megawatt, with upfront payments covering 50-60% of associated capex
• Nebius short-duration deals (up to 6 months) are being negotiated at $40-50 million per megawatt and sometimes above — roughly 2x the midterm rate
• Nebius extended its server depreciation schedule this year from four years to five
• Added to SOXX March 2027 puts struck in the low $400s

• Nebius's own disclosure is the short thesis: $40-50M per megawatt for up-to-6-month capacity versus $20-25M per megawatt for 1-3 year contracts is roughly 2x, an inverted term structure in a market where the scarce input — power — does not depreciate.
• Backwardation in compute implies the depreciating asset is the GPU, the customer, or both; Burry explicitly invokes the dot-com precedent that 'customers that would just not quit depreciating is what extended the bust 25 years ago'.
• Management deliberately declining to sell its entire 2027 capacity, despite being able to, is treated as an admission that Nebius itself believes value sits at the front of the curve.
• The accounting contradiction: Nebius's pricing implies roughly 50% annual decay while the company extended server depreciation from four to five years this year.
• The 50-60% upfront capex coverage on the billion-dollar midterm deals is not read as a positive — it is consistent with customers paying to secure near-term delivery rather than committing to long-duration value.
• The short expression is broadening from single names to the index: ORCL at $152 and MU at $924 are single-stock adds, but the SOXX March 2027 low-$400s puts are a defined-risk, longer-dated bet on the whole semi complex.
• The long book is being funded in the opposite direction — MELI at $1,850 and ZTS at $73.60 are non-AI compounders added on the same day the AI trade rallied on CoreWeave's results.

Trading Post August 11, 2026

Burry is pressing the same trade he has been building since November 2025: the AI buildout is now being sustained by credit engineering rather than cash flow, and that is the late-cycle tell. The $500 billion bank-financed vehicle for Nvidia chip purchases matters less for whether it is technically circular than for what it rhymes with — Goldman clearing its $3 billion synthetic CDO in late 2005, which opened the floodgates to the 2006 vintage that destroyed Merrill, and Enron's push to make wholesale power an investable asset class. His deeper accounting work introduces the 'perimeter of debt': hyperscalers and their suppliers pushing obligations off balance sheet and locking in non-cancellable long-term supply contracts, which amplifies the bullwhip on the way down. Layered on top, BTIG's Krinsky notes 182 consecutive sessions without an 80%+ downside volume day — the longest such streak in 30 years — meaning dispersion has masked risk and a correlation-1 selloff has been deferred, not avoided. So he adds to his three largest shorts (expressed through SOXX, NVDA and PLTR puts, PLTR/CAT/ORCL common) while topping up Molina on the long side.

• Added to Molina Healthcare at $198
• Shorted more SOXX at $533
• Shorted more Palantir stock at $175
• Shorted more Caterpillar at $844
• Shorted more Oracle at $145
• NVDA and PLTR puts added at strikes in the low $100s, expiries December 2026, March 2027 and June 2027
• $500 billion deal with Wall Street banks to raise capital for purchases of Nvidia chips
• Goldman Sachs' $3 billion synthetic RMBS CDS CDO warehoused in late 2005
• 182nd consecutive trading day without an 80%+ downside volume day on NYSE — the longest streak in at least 30 years by nearly 50 trading days (Jonathan Krinsky, BTIG)
• No year in at least 30 years has had fewer than five 80%+ downside volume days
• 'The Blessed Fraud Recurrence' had nearly 41,000 views and ~300 comments

• The Molina add is a sizing correction, not a new thesis — the position had shrunk relative to the book because of his own capital movements, and $198 brings it back on par with his largest positions.
• He explicitly refuses the circular-financing debate on the $500B Nvidia chip-purchase facility: the question is not whether it is circular but that unnatural credit is being structured to prolong momentum late in the bull phase.
• The historical template is late 2005 — Goldman warehousing a $3B synthetic RMBS CDS CDO, nearly cracking, then clearing it and opening the floodgates to the 2006 vintage that put fatal risk on Merrill's balance sheet.
• 'Perimeter of debt' is the new analytical frame for his next article: off-balance-sheet structures plus non-cancellable long-term supply contracts reaching deep into the chain, which amplifies the bullwhip on the downswing. Oracle at $145 is the expression.
• Caterpillar at $844 is the physical-world short on the same view, justified in five words: 'Data center plans are already aging.'
• Palantir is now expressed in both instruments — stock shorted at $175 plus added December 2026 and March 2027 puts at low-$100s strikes — implying he wants convexity and carry-free downside across two expiries.
• The put strikes across NVDA and PLTR cluster in the low $100s with expiries from December 2026 to June 2027, so the positions need a large drawdown within roughly 4-10 months to pay, not merely a de-rating.
• Krinsky's statistic is the market-structure argument: 182 consecutive sessions without an 80%+ NYSE downside volume day is the longest streak in 30+ years by nearly 50 days, and no year on record has had fewer than five such days — dispersion has deferred a correlation-1 selloff, not cancelled it.
• His stated survival rule for a cycle he expects to last years: 'Avoid the leverage, and one is more likely to avoid the folly' — everyone can be right and almost everyone can still go bankrupt.

Trading Post August 10, 2026

With Palantir put implied vol crushed to yearly lows while the stock sits at 69x sales, Burry re-establishes his out-of-the-money put position on top of the remaining half of his common short, arguing PLTR's true economics (employee share delivery ~6x reported SBC, tripling off-balance-sheet infrastructure commitments) make the shares worth under $1 long-run. He funds patience on the long side with washed-out quality — adding Zoetis in the low $70s, now cheaper than leveraged competitor Elanco despite far better margins and pipeline. Market-wide, he sees a 1987-like setup at much higher valuations, keeps QQQ and SOXX puts for a step-function break, and expects that when 'price matters again' fallen value names will be the ones that rise.

• PLTR back up to 69x sales (range cited: 50-100x)
• PLTR GAAP net income ~$3.0B over the last year vs 31.3M shares delivered to employees worth ~$5B — about 6x the GAAP stock-based-comp expense, the widest gap of the 66 companies he has worked up post-earnings
• PLTR off-balance-sheet non-cancellable infrastructure purchase commitments more than tripled this year
• Burry's long-run value estimate for PLTR: under $1 per share
• ZTS has traded 8.3x its shares outstanding (9x+ float) since the 2021 peak near $250; 5.1x float since late 2024; 2.8x float since late 2025
• Elanco: 22% of EBITDA goes to interest expense vs 4% at Zoetis; ELAN ~15x EBITDA / ~20x earnings vs ZTS ~10x / ~12x
• ZTS US segment: 80%+ gross margin and 66% operating margin vs Elanco's projected 55-56% gross margin
• ZTS lost 10 points of US in-clinic dermatology share YoY; Elanco captured maybe 1/3 of the lost dollars
• ZTS $2B December 2025 convertible at 0.25% through June 2029, $148.20 conversion; dilution-hedge calls struck $148.20, capped $211.72, cost $187M
• Zoetis retains a ~2:1 research-dollar advantage over Elanco

• Vol timing drove the PLTR re-entry: he waited until implied volatility collapsed and March 2027 low/mid-$100 strike puts hit their cheapest levels of the year before re-establishing, rather than chasing when the stock first re-rated to 69x sales.
• His post-earnings screen of 66 companies found PLTR has the widest gap between GAAP SBC expense and actual value of shares delivered to employees: 31.3M shares worth ~$5B against ~$3.0B GAAP net income — real dilution running ~6x the reported expense.
• Palantir's off-balance-sheet non-cancellable infrastructure purchase commitments more than tripled this year — part of a broader 'expanded liability perimeter' he says makes today's off-BS liabilities worse than 2000 and 2008.
• Share-turnover as a bottom signal for ZTS: 8.3x shares outstanding traded since the 2021 peak and 2.8x float just since late 2025 means the shareholder base's weak hands are exhausted — his stated rationale for sizing up in the low $70s.
• The ZTS/ELAN pair has inverted: the historically favored Zoetis now trades at ~10x EBITDA vs Elanco's ~15x, despite Elanco paying 22% of EBITDA in interest vs Zoetis's 4% — and in a price war Zoetis's 80%+/66% US margins beat Elanco's 55-56% gross margin.
• A new short target is forming: he is trying to buy his first-ever out-of-the-money Caterpillar puts but refuses to cross wide spreads or pay up for implied vol — watch for execution in a future post.
• He sees 1987 with today's valuations: no imminent prediction, but QQQ and SOXX puts held for a step-function break, with a recovery he expects would be drawn out like post-2000 given record index valuations.
• The regime call: 'Today, price does not matter' — echoing late-1998-to-2000 — and when price matters again he expects the out-of-favor cheap stocks (the field 'from which many stocks will rise') to double or triple as in 2001-2005.

Trading Post August 7, 2026

Burry uses post-earnings and post-downgrade selloffs as his entry mechanism, adding to five longs where the narrative has deteriorated faster than the economics. The clearest expression is Zoetis: the market is treating lost U.S. dermatology share as terminal decline, but Zoetis lost 10 points of share to a fragmented field and category contraction rather than to Elanco alone, and it enters any price war with 80%+ gross and 66% operating margins against Elanco's 55-56% gross margins and a balance sheet consuming 22% of EBITDA in interest. Management's refusal to timeline U.S. stabilization is what created the discount, since Wall Street will not wait 6 to 18 months. On the GSEs, he argues the absence of IPO news is itself the opportunity — Fannie and Freddie compound book value at double digits while the IPO slips, so delay depresses sentiment without destroying value. Against those longs he is pressing a semiconductor short, adding both outright SOXX and March 2027 low-$400s puts specifically because the index's sharp bounce off recent lows improved the entry.

• Added Freddie Mac (FMCC) at $5.43
• Added MercadoLibre (MELI) at $1,812.39, the day after earnings
• Added lululemon (LULU) at $127.45
• Added Fiserv (FISV) at $51.93
• Added Zoetis (ZTS) at $72.72; Zoetis is described as a large position
• Shorted more SOXX at $541 and added to March 2027 SOXX puts struck in the low $400s
• Elanco pays ~22% of EBITDA in interest expense vs ~4% at Zoetis
• Elanco ~15x EBITDA / ~20x earnings vs Zoetis ~10x EBITDA / ~12x earnings
• Zoetis lost 10 points of U.S. in-clinic dermatology share year over year; Elanco captured only about one third of those lost dollars
• Zoetis U.S. segment: 80%+ gross margin and 66% operating margin vs Elanco gross margin guided to 55-56%
• Zoetis $2B December 2025 convertible note at 0.25% through June 2029 with a $148.20 conversion price; capped calls struck $148.20 and capped at $211.72 cost $187M booked to paid-in capital
• Zoetis repurchased stock at $124 with convert proceeds and a lot more at $90 earlier this year

• The Zoetis add is a valuation-inversion trade: the historically premium name now trades at ~10x EBITDA / ~12x earnings while Elanco, the former value half, trades at ~15x / ~20x, with Zoetis converting free cash flow better and carrying far less debt.
• He quantifies the Zoetis bear case to defuse it — 10 points of U.S. in-clinic dermatology share lost, but Elanco took only about one third of those dollars; the rest went to small competitors and category contraction.
• Any Zoetis-Elanco price war favors Zoetis structurally: 80%+ U.S. gross margins and 66% operating margins against Elanco's guided 55-56% gross margins, plus 4% vs 22% of EBITDA going to interest.
• The GSE thesis is that delay is not damage: with no IPO, Fannie and Freddie keep building capital and shrinking ERCF shortfalls while compounding book value at double digits, so a news lull depresses sentiment without impairing value — exactly the setup he flagged in December 2025.
• He is explicit that GSE privatization is 'neither an automatic nor a necessarily quick win' for common and junior preferred holders, which is why he treats weakness as an accumulation window rather than a catalyst trade.
• The Fiserv add leans on seasonality and a transition-year framing, with Mizuho's same-morning call that the bottom is in and that a second 2026 guidance cut isn't needed — Clover strength carrying the Merchant two-year stack.
• His capital-allocation critique of Zoetis is specific: buybacks at $124 funded by the $2B / 0.25% December 2025 convert (conversion $148.20) were too aggressive, and the $187M of capped calls struck $148.20 / capped $211.72 never touch the income statement.
• The semiconductor short was pressed on strength, not weakness — more SOXX shorted at $541 and more March 2027 low-$400s puts bought, with the hard bounce off recent lows cited as the reason the entry was attractive.

Trading Post 8/6/2026: Two Shorts, 3 Earnings Reports with 1 Add & 2 Holds

Burry argues that a cluster of AI-infrastructure companies has quietly financed its build-out with off-balance-sheet liabilities — backstops, uncommenced leases and purchase commitments — that are excused today as evidence of future growth rather than counted as obligations. Because those obligations are enormous relative to the businesses carrying them (he cites uncommenced leases running as long as 11 years at 18x a company's entire sales), the apparent scale and momentum of these firms is borrowed rather than earned, and the same disclosure pattern repeats across the group. That makes them, in his phrase, fish in a barrel: large, similar-looking and short-able almost interchangeably, with the added expectation that sheer size will suffocate them. He acts on it by shorting Oracle at $144.63 and Nebius, in larger size, at $211.77 — common stock rather than puts, since implied volatility above 100% makes the options unpriceable. On the long side he does the opposite thing: he uses post-earnings noise in Fiserv, Zoetis and MercadoLibre as an opportunity, treating transitional quarters and credit-portfolio anxiety as reasons to add or hold quality franchises at better prices.

• Shorted Oracle (ORCL) at $144.63
• Shorted Nebius (NBIS) at $211.77, in larger size than ORCL
• NBIS puts rejected: implied volatility greater than 100%
• Uncommenced leases as long as 11 years, totalling 18x the company's entire sales
• FISV $52 vs $48 average purchase price; 84% of revenue still growing; Clover revenue +13%
• ZTS $78 vs $83 average purchase price; trading below IV15
• MELI $1,782 vs $1,611 average purchase price; target buy zone $1,500 or less, hoped-for range $1,300-1,400
• MELI net revenue surpassed $10 billion for the first time, +50% y/y
• MELI income from operations $683 million, 6.7% margin
• MELI credit portfolio grew 75% last quarter; credit-card users 2-3x more likely to become 'ecosystemic'
• Position sizing: ADBE and JD largest; next cohort includes ZTS, FLUT, FISV; MELI a step below

• Burry opened two fresh AI-infrastructure shorts on the same day — Oracle at $144.63 and Nebius at $211.77 — with Nebius in the larger size, marking the first time the off-balance-sheet thesis has been expressed as live short positions rather than research.
• The analytical hook is uncommenced leases: he flags a company with lease commitments running up to 11 years and totalling 18x its entire annual sales, obligations that sit outside the balance sheet and get read as growth rather than debt.
• He explicitly says this is not a single-name problem — the companies 'all look similar and probably any single one would do' — implying the short list is longer than the two names disclosed, with Part IV of the Heretic's Guide to follow.
• He rejected Nebius puts despite the bearish conviction: implied volatility above 100% made option prices nonsensical, so he took the equity short and the unlimited-loss profile that comes with it.
• Fiserv is a deliberate turnaround bet, not a growth bet: he adds at $52 versus a $48 average and treats the print as the new CEO's kitchen sink, leaning on the 84% of revenue still growing and Clover's +13% as evidence the franchise is intact.
• Zoetis is being held through a mark-to-market loss ($78 vs $83 average) on the argument that soft North American pet demand is cyclical and known, with the shares below IV15 and a new outsider CFO/COO from GE HealthCare as a positive change agent.
• MercadoLibre's selloff to $1,782 is, in his framing, investors mistaking deliberate reinvestment for deteriorating quality — $10bn revenue at +50% y/y and a 6.7% operating margin the company says is a chosen trade-off — with price/sales at lows and a stated add level of $1,500 or below.
• The book's shape is now disclosed: ADBE and JD are the largest positions, followed by a cohort including ZTS, FLUT and FISV, with MELI smaller — a long book of unloved quality franchises funded against short exposure to AI-infrastructure balance sheets.

Trading Post August 5, 2026: One Buy, One Sale, One Reason

Prediction markets, led by Kalshi at roughly $39.7 billion of annualized volume, are cannibalizing the U.S. online sportsbooks by self-certifying event contracts with the CFTC and claiming a federal exemption from state gambling law — which lets them sidestep state gaming taxes as high as 51%. That regulatory arbitrage, not a structural loss of consumer demand, is what just collapsed Flutter's U.S. EBITDA by 70% and forced a guidance cut and higher leverage. Burry's view is that a tax loophole of this size will not stand: courts, states hunting the revenue, and eventual federal regulation will restore a level playing field, and the incumbents are already building their own prediction-market products rather than fighting. If that happens, the sell-off in the incumbents is a valuation dislocation rather than a terminal-decline signal. He therefore wants the cheapest, highest-quality asset in the space and only that one — Flutter below his intrinsic-value estimate, not DraftKings above it.

• Flutter U.S. gambling EBITDA down 70% to $119 million
• Flutter International EBITDA up 10% (4% organic) to $2.3 billion
• Flutter U.S. revenue down 6%; sportsbook revenue down 15%; handle up only 2%
• Promotional spend up 140 basis points
• Full-year leverage guided up from 3.7x to 4.3x versus a 2.5x target
• FLUT fell ~$15 to near $90
• Kalshi running $39.7 billion annualized volume, 87% of it sports
• State gaming taxes as high as 51%, which prediction markets avoid via CFTC self-certification
• 64% implied odds the Supreme Court agrees to hear a prediction-markets case by year-end (36% it does not)
• Burry's IV15: $96 for FLUT; DKNG IV15 'much lower than today's price'
• New Flutter average cost in the $98s
• CEO Peter Jackson stepping down after 9 years, succeeded by Dan Taylor

• Flutter's U.S. EBITDA collapse (down 70% to $119 million) sits alongside International EBITDA up 10% to $2.3 billion — the damage is specifically American, which points at a U.S.-only regulatory cause rather than a global demand problem.
• Handle grew only 2% while promotional spend rose 140 basis points: FanDuel is paying more to hold flat volume, the signature of share loss to a cheaper competing venue.
• The guidance cut pushes leverage from 3.7x to 4.3x against a 2.5x target, so the equity is absorbing balance-sheet risk on top of the earnings miss.
• Kalshi's $39.7 billion annualized volume — 87% sports — is the scale at which prediction markets stop being a curiosity and start showing up in incumbent P&Ls.
• The core asymmetry is tax: CFTC self-certification lets prediction markets claim a federal exemption and avoid state gaming taxes of up to 51%, a cost gap no sportsbook can out-market.
• The incumbents have stopped litigating and are instead building their own prediction markets while letting revenue-hungry states carry the legal fight — a cheaper strategy with the same payoff.
• Prediction markets price a 64% chance the Supreme Court takes a case on their own legality/taxation by year-end, which Burry treats as the timing variable for the thesis.
• The trade is a pure relative-value swap, not a sector call: DKNG sold because its IV15 is far below price, FLUT more than doubled because its IV15 of $96 sits above a ~$90 print.
• He expected DraftKings to miss when it reported the next afternoon — the DKNG exit was timed ahead of that print, not after it.

Trading Post August 4, 2026: My Options

The S&P 500 has broken to new highs on a rare four-day 5% surge whose only historical analogs cluster around major tech tops, but Burry argues the mechanics of the rally — falling headline volatility forcing vol-targeting funds to lever up and dragging other momentum strategies in — mean new highs will still pull money in. Underneath, the damage done in July has not healed: the Goldman momentum pair trade remains broken and the VIX/VIXEQ ratio at 12-year lows shows calm index volatility masking violent component-level moves. He therefore keeps the structural short on, concentrated in semiconductors and momentum leaders, while managing the options book around price: rolling out duration where volatility is tolerable and taking profits or letting positions lapse where premium is too rich. Working forensically through the Big Five hyperscalers' earnings and filings has already changed his positioning — most visibly by removing his one bullish expression, the Microsoft calls.

• S&P 500 surged 5% over four trading days to a new all-time high — per Jonathan Krinsky (BTIG) only three prior analogs: 3/21/00, 4/23/99 and 11/9/20
• Equal-weight S&P 500 had already made new highs; the Nasdaq 100 remains some way below its all-time high
• VIX/VIXEQ ratio sits at 12-year series lows
• Goldman Sachs Momentum Pair Trade still damaged after a historic run, with data back to the GFC
• SOXX 30-minute chart over six months shows momentum turning back to the upside
• NVDA puts rolled from December 2026 to June 2027, strikes kept in the low 100s
• QQQ puts rolled from January to February 2027 at a small credit, strikes in the high 500s on both legs
• ORCL January 2027 low-100s puts closed for a substantial profit
• Half the Palantir put position sold; the other half held un-rolled
• Short book (SOXX, MU, NVDA, CAT, PLTR, TSLA, AMAT) profitable in every name except NVDA

• The only three prior instances of a 5% four-day surge into a new S&P 500 high were 3/21/00, 4/23/99 and 11/9/20 — two of them near major tech tops, which is why Burry still frames a 1987-type fall as possible.
• He is explicit that new highs are self-reinforcing in the short run: rising prices on falling volatility force vol-targeting funds to lever up and pull other momentum strategies in.
• The VIX/VIXEQ ratio at 12-year series lows is his key tell that calm index volatility is masking large single-stock volatility underneath — the condition that produced July's damage.
• The Goldman Sachs Momentum Pair Trade remains impaired after a historic run; with data back to the GFC, he points to how an extended pair trade eventually unwound as the template.
• Selling the Microsoft calls removes his only bullish options expression and is a direct output of his forensic work on the Big Five hyperscalers' earnings and filings.
• Option premium — not conviction — is now driving his sizing: he rolled NVDA (to June 2027, low-100s strikes) and QQQ (to February 2027, high-500s strikes, at a small credit) but let the Palantir puts go because replacement vol was too expensive.
• The Oracle January 2027 low-100s puts were closed at a substantial profit, with an explicit re-entry condition: lower implied volatility.
• He holds the CAT and PLTR shorts through strong earnings that lifted both stocks materially, and discloses NVDA is the only name in the short book not currently showing a profit — with a stated stop-loss discipline if the shorts run against him.

[Abridged] Foundations: Market Structure, Volatility Targeting, Pod Shops & Other Gremlins

The condensed version of the July 31 Foundations essay, written for readers who want the mechanism in under 1,500 words. Modern market behavior — ignoring rallies, panicking 2-4x harder on declines, then forgetting the fear twice as fast — is not human, so Burry attributes it to three mechanical forces acting in sequence. Vol-targeting funds, whose exposure is inversely proportional to prevailing volatility, sell convexly and early. Pod shops, forced by risk limits to unwind at many multiples of their actual losses, supply the extra chaos. Both are spent by the time the index is down 3%, leaving a price-insensitive passive bid to repair the damage. The result is a market that structurally cannot get cheap, ratcheting valuation to the 100th percentile of 145 years, and the only thing that would reset it is a drop deep and long enough to make the passive machine reallocate.

• 2011-2026 delivered 12.09% annualized real total return with dividend contribution of only 1.76% — 1st percentile across 145 years
• A big down day today triggers ~4x the chaos of 1985-2000, with ~2x faster decay of residual fear (GJR 1993)
• S&P intraday volatility 13.83% (1985-2000) vs 13.08% (2011-today); close-to-close rose 16.64% to 17.50%, entirely outside exchange hours
• Vol-targeting rule: Exposure = Target Volatility / Prevailing Volatility — vol at 5% implies 200% of assets, vol at 20% implies 50%
• A 2.5% drawdown lifting realized vol to 20% forces a cut from 77% to 50% of assets; a further -4% day trims only 8 more points (33% to 25%)
• Pod shops: $400-500 billion leveraged through ~1,800 pods into several trillion in assets, controlling over 30% of trading volume
• A $500m pod at 6x ($3bn book) losing $25m has half its capital pulled, re-levers at 4x, and dumps $2bn of exposure; 50 such pods need 1.5 days of all index volume to clear
• Morgan Stanley Momentum TMT index fell 53.5% in a week the S&P looked calm
• S&P spent only 1.96% of sessions more than 20% below a prior high over 15 years vs a 17.7% median since 1928; none of 1,215 comparable-drawdown blocks recovered faster
• Mean 15-year CAPE of 29.57 is the 100th percentile of 145 years of Shiller data; his S&P 500 CAPE estimate of 44.8-45.7x clears the 1999 record of 44.20. Stated reset condition here: a 25% loss maintained over 2 years

• This is the free/abridged companion to the July 31 Foundations essay, aimed at delivering the mechanism in under 1,500 words; the analytical spine survives but the supporting datasets and cross-references are thinned.
• The vol-targeting equation is presented as the single most portable takeaway: Exposure = Target Volatility / Prevailing Volatility, with 1/x convexity meaning exposure doubles to 200% at 5% vol and halves to 50% at 20% vol.
• The front-loading point explains the market's odd signature — a 2.5% drawdown forces a cut from 77% to 50% of assets while a further -4% day trims only 8 more points, so the chaos is worse at small declines than large ones.
• The pod amplification example is retained in substance: a $500m pod at 6x turns a $25m loss into $2bn of forced exposure reduction, and 50 such pods would need 1.5 days of total index volume to exit.
• Because pod books are roughly half long and half short, index-level damage is muted while single stocks are destroyed — the Morgan Stanley Momentum TMT index fell 53.5% in a week the S&P looked calm.
• The reset condition differs between versions: the abridged post says a 25% loss maintained over two years, while the full July 31 essay specifies a 25% drop held more than 400 days on the 1987 template.
• Valuation conclusion is unchanged — 15-year mean CAPE of 29.57 at the 100th percentile of 145 years, and an adjusted S&P CAPE of 44.8-45.7x above the 1999 record of 44.20.
• Positioning is stated in one line with no new names: long undervalued out-of-favor securities, short highly valued momentum favorites, plus his standing warning that most people should not short.

Foundations: Market Structure & What The Heck Just Happened

The modern market's behavior cannot be explained by human psychology, so it must be explained by machinery. Running GJR (1993) and Lo-MacKinlay (1988) on 1985-2000 versus 2011-2026, Burry finds a market that ignores rallies entirely, panics with roughly 4x historical intensity on big declines, then forgets that fear twice as fast — and that reverts rather than trends day to day. He attributes the pattern to a village of mechanical actors: vol-targeting funds whose exposure is inversely proportional to prevailing volatility, so they dump convexly and early; pod shops whose risk limits force unwinds at 40-200x their actual losses; and a price-insensitive passive bid that shows up on every decline and repairs the damage. The consequence is not that these players make stocks expensive — none has a view on price — but that they disable the process by which markets used to get cheap, so multiples ratchet up and never reset. He is not calling a top; he is naming the condition for one, which is a drawdown deep and long enough to make the passive machine reallocate.

• S&P 500 delivered 12.09% annualized real total return 2011-2026 with dividend contribution of only 1.76% — 1st percentile across 145 years of 15-year blocks
• A big down day today triggers ~4x the chaos of 1985-2000, but residual fear dissipates ~2x as fast (GJR 1993 framework)
• Two-day Lo-MacKinlay variance ratio is 0.881 (z = -2.56, 3,774 sessions) vs 1.02 in 1985-2000 — the market now walks itself back rather than trending
• Intraday S&P volatility fell from 13.83% to 13.08% while close-to-close rose from 16.64% to 17.50% — the entire increase occurred outside exchange hours
• Vol-targeting capital ~$500 billion against a $60 trillion US market — but it arrives all at once in S&P 500 futures, where the marginal dollar sets price
• Pod shops control ~$428 billion of investor capital (Pensions & Investments, April) leveraged through 1,400-1,800 pods, and can be over 30% of daily trading volume
• A $500m pod at 6x gross loses 5% of capital on a 0.83% adverse move; halved and re-levered at 4x, a $25m loss forces $2bn of turnover — 80x. Nasdaq 100 trades only ~0.23% of shares outstanding a day, so 99.77% sits still
• S&P spent just 1.96% of sessions more than 20% below a prior high over 15 years vs a 17.7% median since 1928; 1987's -33.5% took 414 sessions to a new high vs 103 for 2020's -33.9%
• Mean 15-year CAPE of 29.57 is the 100th percentile of 145 years of Shiller data; his adjusted S&P 500 CAPE of 44.8-45.7x clears the 1999 record of 44.20
• 10-year Treasury at 4.7347%, an 18-month high; his stated reset condition is a 25% index drop maintained more than 400 days

• The dividend contribution to the S&P's 2011-2026 real return was 1.76%, the 1st percentile of 145 years — buybacks and RSU withholding replaced the only non-reversible component of shareholder return with an employee cash transfer invisible to earnings and standard FCF.
• GJR (1993) applied to 1985-2000 vs today shows ~4x the chaos on a big down day but ~2x faster decay of residual fear, while Lo-MacKinlay's variance ratio fell from 1.02 (random walk) to 0.881 with z = -2.56 — the market now mean-reverts rather than trends.
• The entire rise in close-to-close volatility (16.64% to 17.50%) occurred outside exchange hours while intraday volatility actually fell, meaning risk has migrated to the overnight gap.
• Evidence points away from forced deleveraging: 2%+ down days close in the bottom quarter of range only 81% of the time in the 2020s vs ~90% historically, and the effect vanishes in the equal-weighted S&P — so the mechanical bid is concentrated in cap-weighted indices.
• Vol-targeting's convex 1/x exposure rule front-loads the selling: a 2.5% drawdown forces a cut from 77% to 50% of assets while a subsequent -4% day trims only 8 more points, which is why the market is more chaotic at 2-3% declines than at larger ones.
• The pod-shop amplification math is the essay's sharpest number: a $500m pod at 6x turns a $25m loss (0.83% of book) into $2bn of forced turnover — 80x — and fifty such pods out of ~1,800 would need 1.5 days of all Nasdaq 100 volume to exit.
• Repair speed has quadrupled: 1987's -33.5% took 414 sessions to a new high, 2020's -33.9% took 103. Over 15 years the S&P spent just 1.96% of sessions more than 20% underwater vs a 17.7% historical median, and none of 1,215 comparable-drawdown blocks recovered faster.
• He declines to call a top and instead names the condition for a real reset: a 25% drop held more than 400 days, which would clear the pod shops and vol traders and possibly push the passive bid toward 6% ten-year bonds.

Trading Post July 30, 2026: A Sizable Batch of Trades, But Only Two Kinds of Charts

The one-day collapse and immediate snapback in the Momentum Pair trade was a historic reversal — larger, he says, than the analogous episode 26 years ago — and the market's reflex has been to put the trade straight back on. Burry reads the rebound as tired rather than durable, and uses it to press both sides of his book. On the long side he keeps buying out-of-favor names that have spent a long time falling and are now basing, which he interprets as shares turning over into stronger hands and therefore de-risked for long-term holders. On the short side he adds to the semiconductor and AI complex through Micron, SOXX, and dated puts on Nvidia and the Nasdaq 100. The structural argument behind the trades is deferred to a forthcoming piece on hyperscaler earnings quality.

• DKNG added around $23.40
• ZTS added around $76, now a full position
• LULU added around $118, now a full position
• QQQ Jan 15, 2027 puts added in the high $500s strike range
• NVDA Dec 18, 2026 puts added in the low $100s strike range
• MU shorted more at ~$880
• SOXX shorted more at ~$506
• All buys and additions in this post — no sells

• Every trade in the post is a buy or an add — Burry explicitly notes there are no sells, which is unusual for a Trading Post covering a violent two-day market move.
• DKNG at ~$23.40 and FLUT together form one of his larger positions, making sports betting a meaningful long-side allocation rather than a starter.
• ZTS at ~$76 and LULU at ~$118 both reached full-position size on this day; LULU extends the thesis from his June 24 deep dive on the company's management.
• His long-side screen is explicitly technical, not just fundamental: a long fall followed by basing/consolidation, which he interprets as share turnover into stronger hands and therefore a de-risked entry.
• The AI/semis short is expressed four ways at once — MU common at ~$880, SOXX at ~$506, NVDA Dec 18 2026 puts in the low $100s strikes, and QQQ Jan 15 2027 puts in the high $500s strikes — spreading the bet across single name, basket, and index.
• He read the same-day snapback in the Momentum Pair trade as a knee-jerk, calling the reversal more extreme than the analogous 1999-2000 episode and judging the trade's remaining legs "tired."
• The post pre-announces a hyperscaler earnings-quality piece on Microsoft and Meta, flagging "accounting maneuvers and poor earnings quality" ahead of the filings — a signal of where the next short-side work was headed.

Trading Post July 24, 2026 - 4 Shorts, 3 Longs

On a busy trading day Burry bought three longs and expanded four shorts, all expressions of theses he says only work if he stays patient. The longs are a paired bet that regulated sportsbooks Flutter and DraftKings are mispriced by fear of prediction markets, plus an add to managed care in Molina. The shorts extend his central claim that AI demand is not real end demand: per the 2026 BIS annual report, much and possibly most of Nvidia's current and future revenue is financed off-balance-sheet in a circular arrangement, so he added to NVDA and to the broader SOXX complex, plus Micron and Caterpillar. He remains short Tesla, which he says shrinks on its own, QQQ through puts, and Palantir - whose successful protest forcing the DIA to abandon its own fully operational intelligence system he reads as a company that must litigate to keep its customer.