Situational Awareness reported $20.2bn of 13F positions at 30 June 2026; at fixed share counts, its twelve largest positions fell 42.5% by 29 July, while MSCI World slipped 1.6%. At twice equity in long exposure, that fall would consume 85% of equity before financing costs; the fund's actual leverage is not disclosed. CNBC reported on 30 and 31 July 2026 that margin calls forced the fund to sell its public book to Citadel at a discount.
Aschenbrenner's letter, as reported by Hedge Fund Alpha on 31 July 2026, put net July performance at -67% and 2026 at +80%, both interim and unaudited. Copio's assessment: the fund made one concentrated bet on AI infrastructure, financed it with borrowed money, and shorted software in a way that added to the same bet.
The retail lesson is to check whether a shared exposure, borrowing or a spending deadline could force sales during a fall. A hedge earns its place only if its protection improves the whole portfolio's expected compounded growth after costs, or meets a defined spending need.
The adviser, Situational Awareness LP, registered with the SEC on 24 February 2025 and manages a Delaware master fund with a Cayman feeder. Aschenbrenner is managing partner and Carl Shulman co-portfolio manager, according to the Core Scientific Schedule 13D. The Form ADV amendment filed on 11 June 2026 reports a $9.28bn gross asset value, 81 beneficial owners, a $5m minimum and four prime brokers.
Those brokers are BofA Securities, Citigroup, Goldman Sachs and J.P. Morgan. Form D filings for the onshore fund and offshore feeder, signed on 11 September 2026, record $9.53bn of interests sold since November 2024, to 135 and 52 investors respectively. Only $1.96bn had been sold by 10 March 2026, so about four fifths of the cumulative interests sold were reported after that date. These are gross offering totals, not net flows or fund returns, and the September filings do not establish how much arrived before July's losses.
Reported 13F value: $0.25bn at 31 December 2024; $13.68bn at 31 March 2026, of which $8.46bn in puts; $20.24bn at 30 June 2026, of which $20.17bn in shares. Source: Form 13F-HR filings, Q4 2024 to Q2 2026; all 152 filed lines (CSV).
The latest 13F, for 30 June 2026, lists 26 lines. Sandisk ($5.67bn) and Micron ($5.57bn) make up 55.6% of the reported value. Bloom Energy ($1.90bn), TSMC's ADR ($1.27bn), Nebius ($1.23bn) and CoreWeave ($0.74bn) follow, then Core Scientific, STMicroelectronics, Applied Digital, Riot Platforms, SharonAI and IREN. Options had almost disappeared.
Three months earlier the 31 March 2026 filing looked very different. Puts made up 61.9% of its $13.68bn, including $2.04bn on the VanEck semiconductor ETF and $1.57bn on Nvidia. A 13F values options at the underlying shares, so these figures are underlying notional values, not premiums, option market values or delta-adjusted exposure. Adding them to share values does not produce the fund's assets under management. The filing does not say whether the puts hedged the longs or were bets of their own.
The SEC's 13F guidance excludes short positions (FAQ 41), options the manager writes (FAQ 43) and shares listed only outside the US (FAQ 40). So the filings show neither the software shorts nor SK Hynix, which CNBC named among the fund's losing longs on 30 July 2026. CNBC also reported on 31 July that about a third of the fund sat in private companies, dominated by a multibillion-dollar Anthropic stake.
Ownership filings after the unwind show that not every public position went to Citadel. The Core Scientific Schedule 13D/A records two block sales of 5,759,539 shares each on 3 August 2026, at $19.69 and $19.48, leaving 14.1m shares. A SharonAI Schedule 13D shows the fund holding 21.1% of SharonAI's Class A shares after exercising warrants on 15 September.
The portfolio follows the argument of Aschenbrenner's June 2024 essay series, Situational Awareness. He wrote that "many trillions of dollars will go into GPU, datacenter, and power buildout before the end of the decade." He called power "the single biggest constraint on the supply-side" and described HBM memory and advanced packaging as "already key bottlenecks for the current AI GPU scaleup".
The June book maps onto those bottlenecks: memory makers, a power-equipment maker in Bloom Energy, and AI computing capacity in CoreWeave, Nebius, Core Scientific and IREN. His author page describes "an investment firm focused on AGI, with anchor investments from Patrick Collison, John Collison, Nat Friedman, and Daniel Gross." The 13D filings add only the standard statement that the shares were bought because they were "undervalued".
No fetched primary document explains the leverage or the short book. CNBC's report on Jim Cramer's comments, on 30 July 2026, said the fund "embraced the popular 'long hardware, short software' trade", shorting software stocks seen as vulnerable to AI disruption. Hedge Fund Alpha reported that the letter says the fund "was not shut down, liquidated, or transformed into a private-only fund" and now runs its public book unlevered.
Situational Awareness publishes no return series, and none of the filings contains one. The only fund returns in the fetched sources are two interim, unaudited net figures from the July 2026 letter, as reported by Hedge Fund Alpha. Geometric return since inception, Sharpe and Sortino therefore cannot be calculated, and we do not substitute a replica. The table uses 30 July for the interim fund figures, whose exact valuation cutoff is unverified, and the index comparison. The last row is a separate fixed-share price calculation, not a fund return.
| Window | Fund, reported | MSCI World |
|---|---|---|
| 2026 to about 30 July | +80% | +9.7% |
| July to about 30 July 2026 | -67% | 0.0% |
| Jan-Jun 2026 | +445% implied | +9.7% |
| 30 Jun-29 Jul, disclosed longs | -42.5% | -1.6% |
The implied first-half figure is our arithmetic, 1.80 divided by 0.33, not a reported number. It shows how compounding treats such a path. A +445% half-year followed by a -67% month compounds to about +80%; averaging returns over those unequal periods would not give a meaningful performance measure. Recovering from -67% requires +203%; our recovery maths article explains why the rebound is earned on a smaller base.
The 13F value rose 79-fold, from $0.25bn to $20.24bn, between December 2024 and June 2026, while MSCI World rose 32.8%. That multiple is not a return. It can reflect subscriptions, borrowing, option notional values and price changes, and it covers only reportable long positions. The $9.53bn offering total extends to September, beyond this June endpoint, so it cannot reconcile the increase. The reported $45bn peak in assets, attributed by CNBC to a person familiar with the fund, has no matching figure in any filing.
Scenario, 30 June-29 July 2026: equity -42.5% unlevered, -85.0% at 2x, -88.9% at 2x long and 2x short IGV; wiped out at 4x or with an Adobe short. Source: 13F for 30 June 2026 and Nasdaq daily closes; scenario (CSV), price inputs (CSV).
These are scenarios: the book changed during July, and CNBC's 31 July 2026 mention of "reports Situational Awareness used as much as 400% leverage" is unconfirmed. At four times equity, a 25% fall in the longs erases the capital; the basket passed that point at the 16 July close, at -29.6%. The reported -67% for July lies between the unlevered and 2x rows.
The reported margin-driven sale is evidence of a structure vulnerable to interrupted compounding; the available data cannot establish its growth-maximising leverage. The objective here is expected log wealth for the fund's investors over several years, under the financing terms of prime brokers who can call margin. In the first half of 2026 the six largest June longs, retrospectively selected and held at fixed share counts, had an annualised daily volatility of 74%. This is a hindsight basket, not the fund's measured volatility or a forecast.
In the standard continuous-time approximation, log growth at leverage L is the cash rate plus L times the expected excess return, minus half of L squared times the variance. At 74% volatility, 2x beats cash only if expected excess return exceeds 55% a year, and 4x needs 109%. In that simplified model, 4x is growth-maximising at 219% expected annual excess return. It assumes continuous rebalancing, stable volatility and borrowing at the cash rate; gaps, financing spreads and margin calls can make it substantially too optimistic.
The counter-evidence is real. At fixed share counts the six positions rose 287% in the first half, Sandisk alone 858%. Those large realised gains explain the appeal of the thesis, but they do not establish the expected return required by the leverage model. The decisive comparison, the same thesis held with less borrowing after fees and financing costs, cannot be run without the fund's fee terms, exposure and NAV history.
It was not adequately protected against the reported July margin shock. At the last disclosed valuation date before the crash, reported long puts were almost gone: $8.46bn of underlying put notional at 31 March 2026 had fallen to $5m by 30 June. The filing omits written options, expiries and strikes, so this is not a complete measurement of protection.
The software shorts then lost money as the longs fell, according to CNBC. Margin financing turned a price fall into forced selling; CNBC reported that the prime brokers marketed the fund's holdings before the 30 July open.
The timing was costly: the disclosed basket rose 20.8% on 30 July, the day after its July low. The rebound did not last, which weakens the idea that the fund simply sold at the bottom: by 6 November the basket stood 49.4% below its 30 June value. On 6 November Hedgeweek, citing Bloomberg, reported a 33% October loss at VARA, a fund whose approach resembles Situational Awareness.
Our Spitznagel study examines protection designed to pay out in such a fall, and what it costs in ordinary years. A more resilient structure could use less callable borrowing, usable liquidity or protection that pays during the relevant losses. Each has a cost: compare terminal wealth across ordinary and stressed paths, including premiums and financing, before treating any of them as an improvement.
No: the longs and the shorts were one bet on AI infrastructure demand, and the reported private stakes sit in the same theme. The matrices below sign-invert the short legs, so a positive number means the short and long returns tended to move in the same direction, including on losing days. In the first half of 2026, the Adobe short already correlated positively with every long, at 0.01 to 0.29. The IGV software ETF short offset a little of Nebius and CoreWeave, at -0.29 and -0.32.
2 January-30 June 2026, 123 days: longs 0.28 to 0.77 with each other; short Adobe 0.01 to 0.29 with the longs; short IGV -0.32 to 0.02. Source: Nasdaq daily closes; correlation matrices (CSV). The short positions are those reported by CNBC on 30 July 2026; their sizes are not disclosed.
1-31 July 2026, 22 days: longs 0.74 to 0.92 with each other; short Adobe 0.66 to 0.84 with the longs. Source: Nasdaq daily closes; correlation matrices (CSV). Twenty-two observations give wide confidence intervals; the direction, not the second decimal, is the finding.
In July the six longs moved almost as one, and the Adobe short moved with them. Adobe rose 28.5% from 30 June to 29 July while the basket fell 42.5%. CNBC's 31 July account describes the result: the longs and shorts "lost money simultaneously". Our falling-together study shows the same pattern across asset classes in bad months.
Shorting software companies expected to lose from AI adds to the same economic thesis as buying AI infrastructure. The observed joint losses fit that interpretation; the prices alone cannot establish why expectations changed. There is no measured protection benefit here to weigh against the shorts' borrowing costs and potential losses.
Copio does not grade this portfolio in its checker, because single stocks, options, borrowing and private stakes cannot be represented by an ETF replica.
For your own holdings, add direct shares to the same companies inside ETFs, group exposures by the revenue source they depend on, and calculate losses before borrowing. Then check the broker's maintenance margin and your spending commitments against that loss. A smaller loan or genuinely different return source needs a comparison of growth after costs; a calmer chart alone is not a reason to adopt it.
Our ARK study examines a concentrated technology fund without borrowing.
Checked on 9 November 2026. Filings come from SEC EDGAR for Situational Awareness LP, CIK 2045724: seven 13F-HR filings for Q4 2024 to Q2 2026, Schedules 13D, 13D/A, 13G and 13G/A for Core Scientific, SharonAI and Nebius, and Forms 3 and 4. The Q3 2026 13F is due by 16 November 2026 and was not yet filed.
Form D filings come from the fund (CIK 2038540) and the offshore feeder (CIK 2048430). The Form ADV is the 11 June 2026 amendment: Item 5.F and Schedule D, Section 7.B.(1). The form does not date its gross asset value. The Part 2A brochure listed for the same date could not be retrieved.
Press figures, attributed to outlet and date and unconfirmed by any filing, come from CNBC on 30 July and 31 July 2026, its explainer and Cramer report, and Hedge Fund Alpha (31 July, updated 19 October 2026). CNBC said on 30 July that the largest holdings at end-March included Nebius and Micron; the March 13F has no Nebius line and Micron mainly as options, so we use the filings.
The basket holds the twelve largest share lines of the 30 June 13F, 96.2% of its value, at fixed share counts and unadjusted Nasdaq daily closes. For eleven lines, value divided by shares equals the 30 June close; Applied Digital implies $30.49 against $37.30. Volatility uses the top six lines, 81.0% of the value, over 2 January-30 June 2026, because SharonAI only began trading in February.
Correlations are Pearson coefficients of daily returns, with short legs sign-inverted. The leverage scenarios assume buy-and-hold from 30 June, with equity return equal to the long multiple times the basket return, minus the short multiple times the short asset's return. They exclude financing costs, fees, borrow costs and margin calls.
MSCI World is the net total return USD index (990100), daily levels from MSCI. All prices are nominal US dollars; stock dividends are ignored, so the basket and index do not share an identical return basis.
The 13F values describe disclosed positions at quarter-ends, not returns or investor capital. The fund's NAV history, fee terms, borrowing, short book and private valuations are not public. Every calculated figure can be reproduced from the CSVs linked above with the formulas stated here. This article is educational, not personalised investment advice.
Its 13F filing for 30 June 2026 reported $20.2bn of US-listed positions, 55.6% of it in Sandisk and Micron, with Bloom Energy, TSMC, Nebius and CoreWeave next. The filing omits short sales, options written, non-US shares such as SK Hynix and private stakes. CNBC reported on 31 July 2026 that about a third of the fund was private, dominated by Anthropic.
Hedge Fund Alpha reported on 31 July 2026 that Aschenbrenner's investor letter estimated net July performance at minus 67% and 2026 performance at plus 80%, both interim and unaudited. The article's headline says June; its own summary of the letter says July. The fund publishes no audited return series.
CNBC reported on 30 and 31 July 2026 that losses on AI infrastructure longs and software shorts triggered margin calls from its prime brokers, and that Citadel bought the public book at a discount. The twelve largest long positions disclosed for 30 June fell 42.5% by 29 July at fixed share counts. Borrowing multiplies such a fall in the fund's equity.
Not in July. With the short legs sign-inverted, the correlation of daily returns between a short Adobe position and the six largest longs ranged from 0.66 to 0.84 in July 2026. Adobe rose 28.5% from 30 June to 29 July while the long basket fell 42.5%, so both sides lost together.
Not usefully. The 30 June 2026 filing appeared on 14 August, two weeks after CNBC reported the public book sold to Citadel. A 13F also leaves out the shorts, the borrowing, non-US shares and private stakes that drove the fund's result. Copio does not grade this portfolio in its checker for the same reason.
