Leopold Aschenbrenner Fund Collapse: What Happened

Last updated: August 5, 2026, 3:30 a.m. EDT

Leopold Aschenbrenner’s hedge fund did not simply make a bad call on artificial intelligence. It encountered a more basic problem: a portfolio can be directionally intelligent and financially fragile at the same time.

Situational Awareness, the AI-focused investment firm founded by the former OpenAI researcher, reported a net gain of 439% through June 30, 2026. Then its portfolio value fell about 67% in July, according to an investor letter reviewed by Reuters and The Wall Street Journal. The firm sold the bulk of a public-stock portfolio described by Reuters as roughly $16 billion to Ken Griffin’s Citadel, removed leverage from the remaining portfolio and acknowledged that it had come uncomfortably close to permanent capital impairment.

The arithmetic is startling but internally consistent. A portfolio that rises 439% turns $1 into $5.39. A subsequent 67% decline leaves about $1.78. That still represents a gain of roughly 78% from the start of the year, close to the approximately 80% year-to-date return reported in the investor letter. The episode therefore cannot be summarized accurately as either a complete wipeout or an ordinary losing month. It was an extreme drawdown that forced a rapid restructuring even though the fund remained substantially ahead for the year.

That distinction matters because several of the most repeated figures describe different things. A reported peak scale of about $45 billion, a March 31 Form 13F total of $13.68 billion, a later public-equity book of roughly $16 billion and a retained portfolio of around $10 billion are not interchangeable measures. They may refer to net assets, gross exposures, reportable securities, private holdings, derivative notionals or assets remaining after sales. Treating the difference between two such figures as a cash loss would be misleading.

The more useful question is how a fund with one of the strongest first-half returns in modern hedge-fund history could become a forced seller within weeks. The answer lies in the interaction of leverage, concentrated positions, daily market prices, prime-broker financing, crowded trades and liquidity that disappears precisely when a large investor needs it most.

Key Takeaways

• What happened: Situational Awareness’s portfolio value fell about 67% in July 2026 after a 439% net gain through June, leaving the fund approximately 80% ahead for the year.
• What was sold: Reuters reported that the fund unwound most of a roughly $16 billion public-equity portfolio and sold the bulk of those holdings to Citadel. The management firm itself was not acquired, and important private investments were retained.
• Why leverage mattered: Borrowing and derivatives magnified gains, but falling collateral values and tighter financing terms could force sales. Those sales could then push down the same securities supporting the loans.
• What public filings show: The fund’s March 31 Form 13F listed $13.68 billion of U.S.-reportable long securities and long option positions. It did not reveal short sales, written options, most swaps, private companies, foreign ordinary shares, cash or the fund’s net leverage.
• Why the story is broader than one fund: Leveraged and inverse exchange-traded products, concentrated AI trades and elevated hedge-fund leverage can create mechanical buying on the way up and mechanical selling on the way down.
• What is not yet known: Public information does not establish the fund’s exact peak leverage, all financing terms, complete July position book, transaction discounts or the contribution of each security to the drawdown.

What Happened to Situational Awareness?

Situational Awareness was built around a thesis that artificial intelligence would require an industrial buildout far larger and faster than conventional forecasts assumed. The firm invested across the physical and financial infrastructure of that buildout: advanced semiconductors, memory, data centers, power equipment, energy supply and companies capable of converting existing assets into AI-compute capacity. It also expressed negative views on businesses that could be disrupted by increasingly capable software models.

For much of 2025 and the first half of 2026, that framework worked spectacularly. Public and private AI-related assets rose, financing was available, volatility often rewarded aggressive positioning and the fund’s reputation attracted additional capital. Returns above 1,000% since inception were reported before the July reversal. The portfolio’s success reinforced both the investment thesis and confidence in the way it was implemented.

Then the market path changed. Semiconductor and AI-infrastructure shares suffered sharp, uneven declines. Memory stocks that had benefited from scarcity, pricing power and intense demand expectations became vulnerable to profit-taking and valuation compression. Software shares that had been treated as likely AI losers rebounded in parts of the market. A long-AI-infrastructure, short-software portfolio could therefore lose on both sides: the longs fell while the shorts rose.

Losses alone do not force a hedge fund to sell everything. Financing constraints do. A fund that owns securities without borrowing can generally decide whether to hold through a decline. A leveraged fund has another decision-maker in the room: the lender. Prime brokers calculate collateral values, haircuts, concentration charges, liquidity add-ons and stress losses. If the portfolio no longer supports the amount borrowed, the fund must post more cash, reduce positions, transfer assets or negotiate new terms.

Reuters reported that Situational Awareness faced pressure to sell assets or raise capital. Goldman Sachs and JPMorgan were among the prime brokers that helped facilitate the sale, according to the news agency. The transaction was assembled quickly, with Citadel emerging as the buyer of most of the public-stock portfolio.

The speed was part of the problem. A portfolio may have an attractive estimated value if sold gradually to many buyers. A large block that must be transferred within a day or two is worth less because the buyer assumes execution risk, market risk, hedging costs and the possibility that other traders will detect the seller’s distress. The discount is the price of immediacy.

Aschenbrenner’s July letter, as reported by Reuters, described the dynamics as similar to a bank run: vulnerability created additional vulnerability. That analogy is imperfect because a hedge fund is not a deposit-taking bank, but the feedback mechanism is recognizable. Falling assets weaken the balance sheet. Weakness changes counterparties’ behavior. Counterparty caution creates a need for liquidity. Raising liquidity through sales can depress the portfolio further.

By the end of the process, the public portfolio had been radically reduced and leverage removed. The fund retained private investments, including its exposure to Anthropic, according to multiple reports. It continued operating. “Collapse” became the dominant shorthand because of the scale and speed of the drawdown, but “forced unwind” is the more precise description of what had been confirmed as of August 5.

The 439% Return Was Also a Risk Signal

A 439% net return in six months is not simply a conventional return delivered faster. It usually indicates some combination of extraordinary asset appreciation, concentrated exposure, optionality, leverage or all four. The result may reflect genuine insight. It also tells investors that the portfolio is capable of moving by amounts that ordinary diversification models do not contemplate.

Exceptional gains can disguise fragility because almost every risk metric improves during a favorable trend. Collateral values rise. Gross exposure becomes easier to finance. Realized volatility may look manageable relative to gains. Lenders see a thicker equity cushion. Investors add capital. Redemptions are limited. Positions become larger in dollar terms even without active buying.

The same process can reverse. A portfolio that compounds upward becomes more concentrated in its winners unless it trims them. If managers use the increased equity to borrow more, the fund may maintain a constant leverage ratio while dramatically expanding its dollar exposure. What looks like stable leverage in percentage terms can therefore become a much larger market footprint.

Suppose a fund begins with $1 billion of investor equity and maintains four dollars of gross exposure for each dollar of equity. It controls $4 billion of positions. If the portfolio rises 25%, the positions become worth $5 billion before financing costs and hedges. Equity rises from $1 billion to $2 billion because the borrowing remains about $3 billion. If the manager restores the portfolio to four-times gross exposure, the fund can expand to $8 billion. A successful quarter has doubled equity but may have doubled market exposure again.

This is why the most dangerous moment in a leveraged strategy can arrive after its best period. The track record attracts capital, lenders compete for business and position limits are relaxed. The manager’s confidence is supported by evidence. Risk controls calibrated to historical volatility may not capture what happens when everyone tries to exit the same assets.

The 439% gain was therefore both an achievement and a warning. It showed that the portfolio contained nonlinear exposures capable of producing several years of conventional equity returns in a few months. The July decline revealed the other half of that distribution.

How the AI Thesis Became a Portfolio

Aschenbrenner’s public intellectual foundation was his June 2024 essay series, Situational Awareness: The Decade Ahead. It argued that continued scaling in computing power, algorithmic efficiency and the removal of practical limitations from AI systems could produce human-level or stronger capabilities within a few years. It further argued that training and deploying those systems would require enormous investments in data centers, electricity generation, chips and supporting industrial capacity.

The investment implication was not limited to buying the largest AI software company. It was to identify bottlenecks. If demand for computing grows faster than supply, the assets controlling scarce inputs can gain pricing power. Those inputs include graphics processors, high-bandwidth memory, networking, power generation, electrical equipment, cooling, data-center land and access to transmission infrastructure.

That bottleneck framework has genuine economic logic. AI systems are digital products, but their production function is physical. A model cannot be trained without chips; chips cannot run without memory and networking; data centers cannot operate without power; power projects cannot connect without transformers, switchgear, permits and grid capacity. A software forecast can therefore become a commodity, industrial and infrastructure portfolio.

The thesis also creates a natural short book. If AI automates parts of programming, customer service, design, research or administrative work, some established software vendors may face lower prices, weaker renewal power or competition from systems built directly on general-purpose models. A manager can attempt to finance long positions in AI beneficiaries by shorting potential victims.

The difficulty is that a sound long-term theme does not specify the path of security prices. A company can benefit from AI demand and still be a poor investment at a sufficiently high valuation. A supplier can report record revenue and fall because the market expected even more. A software company can face long-run disruption and rally for months because earnings are better than feared, short positioning is crowded or investors rotate out of hardware.

In other words, the technology thesis operates on one clock and the financing structure on another. Data-center demand may unfold over years. Margin is calculated daily and sometimes intraday. A fund can be right about 2030 and unable to finance the position through next Tuesday.

What the SEC Filing Revealed—and What It Could Not

Situational Awareness’s Form 13F for the quarter ended March 31, 2026 provides the clearest public snapshot of part of the portfolio. The filing listed 42 entries with an aggregate reported value of approximately $13.68 billion. It included common shares and long options tied to semiconductor, data-center, energy and digital-infrastructure companies.

Among the disclosed common-stock positions were Bloom Energy, CoreWeave, Sandisk, Applied Digital, Core Scientific, IREN, CleanSpark, Riot Platforms, Bitfarms and T1 Energy. The filing also listed long call or put positions referencing Micron, Nvidia, AMD, Broadcom, Oracle, Taiwan Semiconductor Manufacturing and VanEck’s Semiconductor ETF.

Those entries should not be read as a simple list of bullish and bearish bets. Form 13F values options using the market value of the underlying shares represented by the contracts, not the premium paid or the position’s economic delta. A listed put with a notional value of $1 billion may have cost far less than $1 billion and may have served as a hedge, a directional short expression or one leg of a larger structure.

More importantly, the SEC explicitly states that Form 13F does not include short stock positions and does not subtract shorts from longs. Written options are not reported. Many swaps and other over-the-counter derivatives are absent. Foreign ordinary shares that do not trade on U.S. exchanges are generally outside the filing. Private companies, loans, cash and financing liabilities are also missing.

The filing was dated May 15 and covered positions as of March 31. By the time the portfolio came under pressure in late July, almost four months of trading, price movement, fundraising and financing changes had occurred. The 13F is therefore evidence of themes and scale, not a reconstruction of the July book.

The portfolio was more complex than a list of AI winners

The filing’s mix of long shares, calls and puts suggests a portfolio that was not simply “long every AI stock.” Large put positions referenced broad semiconductor exposure and individual companies. Common shares concentrated in businesses tied to power, memory, compute and data-center conversion. This is consistent with a manager attempting to separate favored bottlenecks from securities considered expensive or vulnerable.

It also demonstrates why a portfolio can lose even when it contains hedges. A hedge protects only against the risk it was designed to offset. A broad semiconductor put may not fully cover concentrated long positions in smaller companies. Options can expire, change sensitivity as prices move or lose value when implied volatility falls. A short software book can become a source of loss if software rallies while hardware falls. Cross-asset relationships observed during normal markets can break under stress.

The public record also cannot reveal whether apparently offsetting positions were held in the same fund, the same share class or the same financing account. Hedge-fund organizations may manage multiple vehicles with different liquidity terms and exposures. Without the private books, any claim to know the exact net position would be speculation.

The Mechanics of a Forced Deleveraging

A leveraged portfolio rests on an agreement between the fund and its financing counterparties. The fund supplies equity and collateral. Prime brokers provide cash, stock borrowing, derivatives, custody, execution and balance-sheet capacity. The arrangement works while the value and liquidity of collateral remain sufficient under the broker’s models.

Those models are not static. A large, liquid stock may initially receive a modest financing haircut, meaning the broker lends against most of its market value. The haircut can increase when the security becomes more volatile, when the position becomes a large share of the fund, when daily trading volume falls or when many clients own the same trade. A position can therefore create a margin call even without another price decline: the broker may simply decide that yesterday’s collateral is riskier today.

Concentration is especially important. A $500 million position in a mega-cap stock may be liquid relative to normal volume. A similar position in a smaller data-center, energy or crypto-mining company can represent days of trading. The screen price shows where the last small transaction occurred; it does not guarantee that a large block can be sold there.

When collateral falls, a fund generally has several choices:

• post additional cash or Treasury securities;
• sell liquid assets and use the proceeds to reduce borrowing;
• close short positions or derivatives to lower gross exposure;
• raise new investor capital;
• transfer positions to another counterparty;
• negotiate a portfolio sale to a buyer with enough capital to absorb it; or
• accept liquidation by the broker if no consensual solution is reached.

Each option becomes less attractive as time shrinks. New investors demand favorable terms when they know the fund needs money. Selling the most liquid positions can leave the remaining portfolio harder to finance. Closing shorts may lift the stocks being bought back. Selling longs may depress the collateral supporting the rest of the book. A portfolio transfer can fail if another bank calculates the same risk and refuses the exposure.

The reports on Situational Awareness suggest a negotiated block sale prevented a more disorderly liquidation. Citadel had the capital, trading infrastructure and risk appetite to evaluate a complicated portfolio quickly. Prime brokers had an incentive to facilitate the transfer rather than liquidate positions piecemeal into a falling market. Situational Awareness gained certainty and removed leverage, but likely surrendered some value in exchange for speed and finality.

Why a large fund becomes visible

Large investors rarely trade anonymously in an economic sense, even when their names are not attached to each order. Dealers observe financing requests. Banks see collateral and stress tests. Market makers detect unusual options activity. Traders notice persistent order flow in securities with limited liquidity. Public 13F filings reveal old positions, and social media rapidly compares those filings with current price moves.

Once the market suspects a forced seller, other participants have little reason to provide generous prices. Some step away because volatility and inventory risk are too high. Others hedge by selling correlated securities. Opportunistic buyers wait for a larger discount. Short sellers may press positions they expect the distressed fund to sell. None of this requires a conspiracy or knowledge of every holding. It is enough to recognize that a price-insensitive seller has limited time.

This is the “wounded shark” dynamic Michael Green described in the Prof G Markets discussion. The metaphor is vivid, but the mechanism is ordinary market microstructure. A distressed fund loses the ability to choose when and how to trade. Its counterparties and potential buyers gain that option.

How losses can appear on both the long and short sides

A long-short portfolio is often described as hedged, but “hedged” does not mean immune. Consider a fund that owns AI-infrastructure stocks and shorts established software companies. It expects the first group to outperform the second. The portfolio can make money if both groups rise, provided the longs rise more. It can also make money if both fall, provided the shorts fall more.

The worst relative-value outcome is the reverse: infrastructure longs fall while software shorts rise. Gross exposure matters more than net exposure in that environment. A portfolio can be close to market-neutral by subtracting shorts from longs while still having enormous dollars at risk on each side.

For example, a fund with $250 of longs and $150 of shorts against $100 of equity has $400 of gross exposure and $100 of net long exposure. Its net exposure is only 100% of capital, but gross leverage is four times. If the longs fall 10% and the shorts rise 10%, the fund loses $25 on the long book and $15 covering the short book—a $40 loss on $100 of equity. The market itself may appear only modestly changed while the fund loses 40%.

That is why descriptions such as “four-times leverage” require a definition. Four-times gross exposure is different from borrowing three dollars to own four dollars of one asset. Derivative notional exposure is different from balance-sheet borrowing. Delta-adjusted option exposure is different from the underlying value reported on Form 13F. Public reporting does not establish which measure best described Situational Awareness at the peak.

Leverage Math: How a Small Decline Becomes a Large Equity Loss

The simplest case is a long-only portfolio financed with debt. Start with $100 of investor equity. Borrow $300 and buy $400 of securities. Gross assets are four times equity.

If the securities rise 10%, they become worth $440. The debt remains $300, so equity rises to $140. A 10% asset gain produced a 40% equity gain before interest and fees.

If the securities instead fall 10%, they become worth $360. After subtracting $300 of debt, equity is $60. A 10% asset decline produced a 40% equity loss.

The leverage ratio is no longer four times. It has jumped to six times because the fund now holds $360 of assets against $60 of equity. To return to four-times exposure, it must reduce assets to $240. That requires selling $120 of securities and repaying the same amount of debt. The fund is forced to sell after the decline, not because its long-term view changed, but because its capital base shrank.

A 20% decline is more severe. The assets fall from $400 to $320, leaving only $20 of equity after debt. The fund is now leveraged 16 times. Restoring four-times exposure would require shrinking assets to $80—selling $240 of the remaining $320 portfolio. A 25% decline eliminates the original $100 of equity entirely in this simplified example.

Real hedge funds use shorts, options, swaps and multiple financing arrangements, so the exact path is more complicated. Some positions gain when markets fall. Options have nonlinear payoffs. Brokers may demand margin before book equity reaches zero. Still, the core lesson holds: the equity cushion absorbs the full change in asset value, and leverage makes the percentage change in equity much larger than the percentage change in the portfolio’s holdings.

Illustration assumes a long-only portfolio, fixed debt and no interest, fees, hedges, taxes or margin changes. It is not a reconstruction of Situational Awareness’s actual portfolio.

Why deleveraging can push prices further

After a decline, a leveraged investor must often sell precisely the assets that have fallen. If the positions are large relative to trading volume, those sales move prices. Lower prices reduce the value of the remaining collateral, producing another margin call and another round of sales.

This is an endogenous feedback loop: the portfolio’s risk management creates market flow even without new information about the companies. The decline may begin with a fundamental catalyst, a valuation reset or ordinary profit-taking. Leverage changes the shape and speed of what follows.

The loop can spread to investors who never borrowed from the same prime broker. A competing hedge fund may own the same stock. A market maker may hedge options by selling shares. A risk-parity or volatility-targeting strategy may reduce exposure as realized volatility rises. Retail holders may sell after stop-loss levels are reached. Each participant follows its own rules, but the aggregate flow points in the same direction.

This does not mean every sharp decline is caused by leverage. Fundamentals, earnings, monetary policy and valuation still matter. It means leverage can turn a reassessment into a liquidation event.

Leveraged ETFs Use a Related—but Not Identical—Mechanism

A leveraged exchange-traded fund generally promises a multiple of the daily return of an index or single stock. A 2x bull fund seeks approximately twice the benchmark’s move for one trading day. A 3x inverse fund seeks approximately three times the opposite daily move. The objective resets at the close.

That daily reset is essential. It allows the fund to maintain its stated exposure, but it also means long-term returns depend on the sequence of daily gains and losses. A leveraged ETF is not designed to deliver exactly two or three times the benchmark’s cumulative return over a month or a year.

FINRA has warned for years that compounding can cause results over longer periods to differ significantly from the stated daily multiple, especially in volatile markets. Issuers make the same point in prospectuses and investor education. Direxion, for example, states that leveraged products should not be expected to track a simple multiple beyond one day and require active monitoring by investors who understand the risks.

A simple volatility-drag example

Assume an index rises 10% on day one and falls 10% on day two. An investor starting with $100 finishes with $99: $100 becomes $110, then a 10% decline reduces it to $99. The two percentage moves do not cancel because the second move applies to a different base.

A 2x daily fund rises 20% and then falls 20%. The $100 becomes $120 and then $96. The index lost 1%; the 2x fund lost 4%, not 2%.

A 3x daily fund rises 30% and then falls 30%. The $100 becomes $130 and then $91. The index lost 1%; the 3x fund lost 9%.

A hypothetical 4x daily product would rise 40% and then fall 40%, leaving $84—a 16% loss. The drag grows rapidly with leverage because compounding acts on a larger daily move.

In a smooth, persistent trend, daily rebalancing can work in the investor’s favor. A 2x fund can return more than twice the benchmark over several consecutive up days because gains are compounded on a growing base. The problem is not that leveraged ETFs always decay. The problem is that high leverage combined with high volatility produces severe path dependence, and investors often focus on the destination while ignoring the route.

Why the funds buy after gains and sell after losses

Consider a 2x fund with $100 of net assets and $200 of exposure. If the benchmark rises 10%, the exposure gains $20 and net assets rise to $120. The fund now has $220 of exposure, but it needs $240 to begin the next day at two times leverage. It must add $20 of exposure near the close.

If the benchmark falls 10% instead, the fund loses $20 and net assets fall to $80. Its remaining exposure is $180, but the next day’s target is only $160. It must sell $20.

The fund therefore tends to buy after up moves and sell after down moves. When many leveraged products reference the same stock or sector, their rebalancing can contribute to late-day order flow. Dealers providing swaps may hedge in the underlying securities, transmitting the adjustment into the cash market.

The scale of the effect depends on assets, leverage multiples, daily returns, derivative structures, dealer netting and market liquidity. It should not be assumed that every rebalance directly produces an equal stock trade. Still, the direction is mechanically procyclical.

Why Semiconductors Became the Perfect Stress Test

Semiconductor markets combine several characteristics that make leverage unusually powerful. Demand is cyclical, supply requires years of capital investment, product generations change quickly, and a small number of companies control critical portions of the value chain. AI added a new layer: investors began treating specific chips, memory technologies and infrastructure suppliers as scarce claims on an enormous future buildout.

Memory is especially volatile. DRAM and NAND products have historically moved through sharp cycles because producers make capacity decisions before they know the final level of demand. Prices can surge when supply is tight and collapse when inventory accumulates. High-bandwidth memory used in AI accelerators has different economics from ordinary commodity memory, but it still depends on manufacturing yields, packaging capacity, customer concentration and the pace of data-center spending.

In early 2026, investors gained a more direct way to express the memory thesis. The Roundhill Memory ETF, ticker DRAM, began trading on April 2. Its official materials describe a targeted portfolio of global memory and storage companies, with major exposure to Micron, Samsung Electronics, SK Hynix, Sandisk and Kioxia. Options became available, and Roundhill later introduced RAM, a fund targeting two times DRAM’s daily performance.

The products solved a real access problem. U.S. investors often find it difficult to buy Korean ordinary shares, and broad semiconductor funds include companies from many parts of the chip industry. A memory-specific ETF packaged foreign and domestic holdings into one U.S.-traded security.

That convenience also concentrated flows. When investors bought DRAM shares, the fund or its counterparties had to obtain exposure to a relatively narrow group of companies. If the largest holdings rose, performance attracted additional assets. Leveraged versions and options created more ways to magnify the same theme. The product did not need to be “wrong” to affect the market. It only needed to become large relative to the liquidity of its underlying components.

Michael Green’s July 26 essay, A Semi-Theory of Almost Everything, argued that price-insensitive and leveraged flows helped explain an unusually large portion of the semiconductor move. His central concern was market structure: securities were being bought and sold because a vehicle’s mandate required it, not because a new analyst had changed an estimate of future cash flow.

That argument should be treated as an analytical interpretation, not a settled attribution of every price move. ETF flows can amplify trends, but they interact with earnings, supply forecasts, options positioning, dealer hedging, currency movements, short interest and investor expectations. The correct conclusion is not that fundamentals disappeared. It is that the marginal price can be set by mechanical flow for meaningful periods.

Why memory stocks could fall after strong operating results

A stock price reflects the difference between reality and expectations, not the absolute level of profit. If investors expect a memory producer’s earnings to triple and they merely double, the result can be economically excellent and financially disappointing. If the shares already discount several years of scarcity pricing, even confirmation of strong demand may not be enough.

High valuations also change the reaction to uncertainty. Questions about new Chinese capacity, customer bargaining power, capital expenditures, product delays or the durability of AI demand can have a larger effect after a vertical share-price advance. When leveraged investors own the trade, the first decline can trigger selling unrelated to the size of the fundamental change.

This distinction is central to the Situational Awareness episode. The AI-infrastructure thesis did not need to be disproven for the portfolio to fail its financing test. A temporary 20% or 30% fall in securities that had risen several hundred percent could be enough if gross exposure was large, liquidity was limited and lenders raised margin requirements.

South Korea: When a National Market Meets Leveraged Retail Flow

South Korea became the clearest example of how concentrated technology exposure and leverage can interact. Samsung Electronics and SK Hynix carry enormous weight in the country’s equity market and in household investment culture. A surge in AI-memory expectations helped drive the KOSPI to extraordinary gains in the first half of 2026. The reversal was equally dramatic.

In late July, the benchmark suffered consecutive declines of roughly 10.8% and 6.0%, according to Reuters, while leveraged products and margin financing intensified losses for retail investors. Reports described a peak-to-trough decline near 40% from the June high, with some measures approaching 44% at the intraday extreme.

The policy response was substantial but not a blanket ban on all leveraged ETFs. South Korea announced that individuals would be limited to allocating no more than 20% of total investment assets to single-stock leveraged ETFs. The government raised the minimum deposit for access to 30 million won, introduced simulated-trading requirements, suspended new listings and advertising of single-stock leveraged products, and considered additional market-stabilization mechanisms.

Those details matter because the regulatory debate is not simply “allow leverage” versus “ban leverage.” Authorities can alter who may use the products, how much may be invested, what disclosures are required, how brokers assess suitability, whether advertising emphasizes daily objectives and whether funds may be launched on securities with insufficient liquidity.

South Korea’s experience also illustrates cross-border transmission. A U.S.-listed ETF can create demand for Korean shares. Currency hedges can generate won transactions. U.S. options and swap dealers can hedge in foreign markets or through related instruments. Korean retail investors can simultaneously buy local leveraged products and U.S.-listed securities. The legal wrapper may be domestic, but the risk loop is global.

Did leveraged ETFs crash the KOSPI?

It is too strong to say that one product or product category caused the entire decline. The KOSPI had first risen to valuations and index levels that reflected extraordinary optimism. Earnings expectations were demanding. Memory shares were crowded. Hedge funds and retail investors used margin. Short sellers reacted to weakening momentum. Global AI shares were already under pressure. Those factors would have mattered even without leveraged ETFs.

Leveraged products can nevertheless change the amplitude. Daily-reset funds sell after declines. Margin investors receive calls as collateral falls. Broker risk limits tighten. Dealers hedge options. A market dominated by a few large technology companies has fewer unaffected securities to absorb the shock. The result can be a decline that is faster and less connected to incremental fundamental news than conventional valuation models imply.

The most defensible description is that leveraged ETFs and margin financing acted as accelerants in a market already vulnerable to reversal. They helped translate falling prices into required transactions. They did not create the semiconductor cycle, determine memory demand or single-handedly set every KOSPI price.

Was the AI Trade Broken—or Merely Deleveraged?

A forced unwind can look like a verdict on an investment theme because the largest positions fall together. But liquidation answers a different question from fundamental analysis. It shows that owners needed to sell at prevailing prices. It does not reveal the long-term value of the underlying businesses.

The bullish case for AI infrastructure remains substantial. Large technology companies continue to spend heavily on computing capacity. Training and inference require advanced processors, memory, networking and electricity. The transition from experimental systems to widely used agents could increase demand for data-center services even if individual model providers face intense competition.

Memory is a real bottleneck. High-bandwidth memory packages are technically difficult to manufacture, and qualified supply cannot be created instantly. Data-center power interconnections and electrical equipment also have long lead times. Companies controlling scarce capacity can earn unusually high margins while the imbalance persists.

The skeptical case is equally concrete. Capital spending can run ahead of monetization. Customers may develop more efficient models that use less compute per task. Competition can reduce the price of AI services faster than usage grows. New semiconductor capacity can arrive after the period of greatest scarcity. Companies may fund projects with debt or long-term purchase commitments that look manageable only under optimistic utilization assumptions.

There is also a difference between industry growth and shareholder return. An industry can expand dramatically while investors overpay for the beneficiaries. Railroads transformed the nineteenth-century economy and still produced repeated financial failures. Fiber-optic capacity was essential to the internet, but many telecom investors lost money after the late-1990s buildout. The technology can be transformative while the capital structure destroys value.

The fund’s loss does not prove the AI thesis was false

Situational Awareness reportedly remained optimistic about the fundamentals after removing leverage. That position is logically possible. A manager can conclude that the assets are worth more over time while also admitting that the portfolio was financed in a way that could not survive the path.

Investors should resist two symmetrical errors. The first is to treat spectacular early returns as proof that every forecast is correct. The second is to treat a forced sale as proof that the entire technology thesis is fraudulent. Both confuse price and process with underlying economics.

The more rigorous assessment asks four separate questions:

• Will AI demand grow as quickly as the thesis assumes?
• Which companies will capture the resulting revenue and cash flow?
• What valuation already reflects that growth?
• Can the investor finance the position through plausible drawdowns?

A correct answer to the first question cannot compensate for a failure on the fourth.

From One Fund to the Financial System

The Situational Awareness unwind was large enough to move individual securities, but available evidence did not show a broad failure of major banks or clearing systems. Prime brokers facilitated an orderly transfer. Citadel absorbed assets. The fund removed leverage. Major U.S. indexes recovered quickly, with several returning to records by August 4.

That is the idiosyncratic interpretation: a concentrated fund took too much risk, suffered a severe loss and transferred positions to stronger hands without threatening the banking system.

The systemic interpretation begins with the environment around it. The Federal Reserve’s May 2026 Financial Stability Report said hedge-fund gross leverage was near all-time highs in the latest comprehensive Form PF data and remained concentrated among the largest funds. The Office of Financial Research estimated that the hedge-fund industry had roughly $11.8 trillion in gross assets and leverage of about 2.6 times, with some strategies operating near six times.

High leverage does not automatically imply imminent crisis. Relative-value funds may hold offsetting positions. Treasury arbitrage can involve large notionals and small net risk under normal conditions. Market-neutral portfolios can use substantial gross exposure. The danger appears when correlations change, financing is withdrawn or many funds hold similar trades.

Situational Awareness matters as a demonstration of speed. A fund can move from an exceptional letter to a forced portfolio sale in less than a week. Public data arrive with a lag. Form 13F covered March positions and was filed in May. Form PF data used by regulators are confidential and reported with delays. Other market participants may know a fund is distressed before the broader public understands its exposure.

Why prime brokers are both shock absorbers and transmission channels

Prime brokers reduce risk by demanding collateral, monitoring exposures and forcing clients to deleverage before losses exceed available equity. In the Situational Awareness case, that discipline may have prevented a larger default.

The same actions can transmit stress. If several banks raise haircuts at once, a fund faces simultaneous demands for cash. If multiple funds own the same collateral, their forced sales reduce prices for everyone. If banks compete aggressively during the boom, the system may allow exposures to grow beyond what any single broker observes in full.

The failure of Archegos Capital Management in 2021 demonstrated that problem. Archegos used total-return swaps with several banks, allowing enormous concentrated positions to remain partially hidden across counterparties. When the stocks fell, brokers issued margin calls and sold blocks. Some institutions escaped quickly; others suffered billions of dollars in losses.

Public reporting has not established that Situational Awareness used the same structure or created comparable bank losses. The comparison is about incentives: multiple prime brokers can each believe their loan is adequately collateralized while the client’s combined gross exposure makes an orderly exit impossible.

Historical Comparisons: Similar Feedback Loops, Different Assets

Long-Term Capital Management

Long-Term Capital Management’s 1998 crisis remains the classic example of sophisticated analysis overwhelmed by leverage and crowded positions. LTCM held convergence trades that were expected to profit as price relationships normalized. Russia’s default and a global flight to liquidity caused spreads to widen instead. Positions that appeared diversified behaved similarly under stress because investors sold what they could.

The Federal Reserve Bank of New York organized discussions that led private financial institutions to recapitalize LTCM. The concern was not that every trade was fundamentally worthless. It was that an abrupt liquidation of a huge, leveraged portfolio could destabilize markets and damage counterparties.

The similarity to Situational Awareness is the mismatch between a long-horizon thesis and short-horizon financing. The difference is scale and market breadth. LTCM’s positions spanned major global fixed-income and derivatives markets, while the confirmed 2026 unwind centered more heavily on AI-related equities and private investments.

Archegos

Archegos built concentrated exposures to a small group of media and technology stocks through swaps. The structure increased economic ownership without corresponding public share disclosures and allowed several banks to finance overlapping risk. When ViacomCBS shares fell after an equity offering, margin calls led to block sales that accelerated losses.

Archegos is the more direct comparison because both episodes involved concentrated equities, prime brokers and a race to transfer or liquidate positions. Yet there are important differences. Archegos defaulted and generated major counterparty losses. As of the research cutoff, reports on Situational Awareness described a negotiated sale and continued operation, not a bankruptcy or confirmed default.

The 2018 volatility-product collapse

In February 2018, a spike in the VIX caused inverse-volatility products to lose most of their value in a single session. Some products were terminated. The episode showed how daily rebalancing and derivative hedging can create enormous end-of-day demand in a market that appears liquid under normal conditions.

The connection to leveraged semiconductor ETFs is structural. Products promising a fixed daily exposure must transact when prices move. If their assets become large relative to the underlying market, the rebalancing itself can influence the closing price used to determine the next day’s exposure.

Melvin Capital and the meme-stock squeeze

Melvin Capital’s losses during the 2021 GameStop surge show the opposite side of forced flow. A crowded short can become a source of required buying when prices rise. Short sellers must post more collateral or repurchase shares, and their purchases can drive the price higher.

A long-short AI portfolio can suffer both mechanisms at once: falling long positions require sales while rising shorts require purchases. Gross exposure, rather than a simple net market beta, determines the damage.

What Regulators Can—and Cannot—Fix

Leverage is not inherently abusive. It supports market making, hedging, arbitrage and capital formation. A pension fund may use derivatives to manage duration. An exporter may hedge currency risk. A professional investor may use an inverse ETF for a temporary portfolio hedge. Banning every leveraged instrument would remove useful tools and shift activity into less transparent markets.

The regulatory problem is mismatch: a complex, path-dependent product can be sold as if it were an ordinary long-term ETF; a concentrated fund can obtain financing from several institutions; a security can support leveraged products whose combined exposure is large relative to its trading volume; and disclosures can arrive after the risk has changed.

Possible reforms for leveraged and inverse ETFs

Several targeted approaches are more realistic than a universal prohibition:

• Stronger daily-objective labeling: Product names, brokerage screens and advertising could state prominently that the leverage target applies to one day.
• Suitability and knowledge checks: Brokers could require investors to demonstrate an understanding of compounding, volatility and loss scenarios before trading.
• Concentration limits: Regulators or exchanges could restrict leveraged products tied to single stocks or narrow baskets with insufficient liquidity.
• Exposure caps: South Korea’s 20% portfolio cap offers one model, though enforcement and account aggregation are difficult.
• Dynamic leverage: Products could reduce target leverage when realized volatility or market impact rises, although this changes the product investors purchased.
• Rebalancing transparency: Better estimates of expected end-of-day flows could help markets prepare, but public disclosure might also invite front-running.
• Advertising standards: Marketing could be required to show multi-day loss examples as prominently as amplified upside.

None eliminates path dependence. A daily-reset leveraged product will still react mechanically to market moves. The goal is to make sure investors and market infrastructure reflect that reality.

Possible reforms for hedge-fund leverage

Private funds already report extensive information through Form PF, and prime brokers maintain detailed client data. The gap is aggregation. A lender sees its own exposure but may not know the full portfolio financed elsewhere. Regulators see more of the system but with delays and confidentiality constraints.

Reforms could include more frequent reporting for exceptionally large or fast-growing funds, standardized measures of gross and net leverage, concentration reporting across prime brokers, and stress tests that assume several crowded positions become illiquid simultaneously. Regulators could also examine whether rapid increases in financing should trigger enhanced review even when current collateral appears ample.

These measures carry trade-offs. More public disclosure can expose proprietary strategies and make a fund easier to attack. More conservative margin requirements can reduce liquidity and push activity offshore or into bilateral structures. A rule calibrated to prevent the last crisis may miss the next one.

The strongest case for intervention is not that regulators should decide which AI thesis is correct. It is that financial institutions should understand aggregate leverage and plausible liquidation costs before providing balance-sheet capacity that can affect other market participants.

Why Citadel’s Purchase Matters—and What It Does Not Mean

The buyer on the other side of a forced unwind matters because a large portfolio cannot simply be turned into cash at the last quoted price. A fund that needs to sell billions of dollars of positions quickly faces three practical problems: finding enough buyers, limiting information leakage and preventing its own transactions from pushing prices farther against it. A negotiated block sale can reduce those problems, even when the seller has little bargaining power.

Reuters reported that Citadel acquired most of Situational Awareness’s public-equity positions through a transaction facilitated by prime brokers including Goldman Sachs and JPMorgan. The arrangement appears to have been a transfer of assets rather than a takeover of the fund’s management company. Situational Awareness retained a smaller portfolio that included private investments, and Aschenbrenner remained in control of the firm.

That distinction is essential. Citadel did not necessarily endorse every underlying AI thesis, nor does buying the positions prove that the securities were fundamentally cheap. A large multi-strategy firm can absorb, hedge, finance and gradually exit risk in ways that a stressed concentrated fund cannot. Its decision may reflect the economics of acquiring a portfolio at an attractive discount, the ability to cross exposures against other books, or confidence that it could manage the liquidation more patiently. The buyer’s return profile can therefore be very different from the seller’s, even when both temporarily hold the same securities.

A block transaction also changes the visible market footprint. Instead of placing a long sequence of sell orders into public markets, a fund can transfer risk to a counterparty in one negotiated package. The buyer then decides how quickly to retain, hedge or dispose of the positions. That can reduce immediate disorder, but it does not erase the economic loss. The distressed seller usually accepts a concession for speed, certainty and reduced execution risk.

Prime brokers occupy a central position in this process. They finance positions, hold collateral, calculate margin requirements and monitor the fund’s capacity to meet obligations. When volatility rises, they may require more collateral or reduce the amount they are willing to lend against a position. A fund can be solvent on a long-term fundamental view and still become unable to satisfy short-term financing demands. Once several lenders seek protection simultaneously, the fund’s preferred timetable becomes irrelevant.

The transaction therefore illustrates a feature of modern markets that is easy to miss: a portfolio’s fate can be determined by the liability side of its balance sheet. Analysts often spend most of their time debating whether a stock is overvalued or undervalued. A leveraged investor must also ask whether the financing remains available, whether collateral is liquid, whether counterparties can change terms and whether the portfolio can survive a temporary price move. A thesis can ultimately be correct and still produce a disastrous investment result if the capital structure cannot endure the path.

Four Numbers That Should Not Be Treated as Interchangeable

Coverage of the episode has circulated several large figures: approximately $45 billion, roughly $16 billion, about $13.7 billion and around $10 billion. They describe different concepts and dates. Treating them as a single sequence creates the false impression that one simple subtraction measures the fund’s loss.

1. The reported peak scale of roughly $45 billion

The largest figure has been used to describe the fund’s peak size or portfolio scale. Public reporting does not provide a complete audited bridge showing how much represented investor net asset value, borrowed capital, derivative notional exposure, private-company marks or other gross positions. It should not be read as cash entrusted by clients or as a conventional mutual fund’s assets under management.

2. The roughly $16 billion public-equity book

Reuters described the portfolio being unwound as roughly $16 billion of public equities. That is closer to the package involved in the sale, but it still does not necessarily equal the fund’s net capital. It may include gross exposure financed with debt, options or offsetting positions. It also refers to a later point than the March quarter-end SEC filing.

3. The $13.68 billion Form 13F total

The March 31 filing listed $13.68 billion of reportable U.S.-listed long securities and long options. It excluded short stock positions, many foreign securities, private assets, written options and numerous swaps. Option values in a 13F are based on the market value of the underlying shares represented by the contracts, not the premium paid or the position’s economic delta. The headline total therefore cannot be added to or subtracted from the other figures as though all were measured on the same basis.

4. The approximately $10 billion retained portfolio

Reporting after the unwind described a smaller remaining portfolio of approximately $10 billion, including private investments. That figure reflects what remained after public positions were sold, not necessarily the amount of equity capital left for investors. Private holdings can be valued less frequently and may not respond immediately to public-market declines. Their apparent stability is partly a feature of valuation timing.

The cleanest independently reported performance measure is the 67% July decline in the investor letter. Even that number needs context. Fund returns depend on the valuation policy for private holdings, the treatment of fees and expenses, the timing of subscriptions and redemptions, and the share class being reported. It is more informative than trying to infer losses by subtracting two differently defined portfolio figures, but it is not a substitute for audited financial statements.

The Exact Leverage Ratio Remains Unclear

The Prof G Markets discussion repeatedly referred to leverage of roughly four times. That may be directionally consistent with a heavily financed portfolio, but no public regulatory filing reviewed for this article establishes a precise four-times ratio for the fund at the moment of the July drawdown. Form 13F does not disclose borrowing, short exposure, swaps or counterparty financing. The investor letter excerpts reported by news organizations also did not publish a complete balance sheet.

There are several ways to calculate leverage, and they can produce different answers. Gross leverage divides total absolute long and short exposure by net asset value. Balance-sheet leverage compares assets with equity. Regulatory leverage may apply conversion formulas to derivatives. A risk model may adjust positions by delta, duration, volatility or expected loss. A fund can therefore be “four times levered” under one convention and materially different under another.

This uncertainty does not invalidate the central mechanism. The fund acknowledged removing all leverage after the loss, and the reported need to sell public positions under pressure indicates that financing constraints mattered. The responsible conclusion is not that a particular multiple has been proven. It is that leverage was material enough to convert a severe market decline into a forced restructuring of the portfolio.

Precision matters because leverage labels can sound more definitive than the underlying data. A long-short fund with 300% gross exposure and 20% net exposure is not equivalent to a directional fund with 300% net exposure. A portfolio of liquid Treasury futures is not equivalent to a portfolio of volatile single stocks at the same notional multiple. A private asset valued quarterly does not create the same daily margin demands as an exchange-traded security. The ratio is only the beginning of the risk analysis.

Risk Management Failed Before the Thesis Was Settled

The most revealing feature of the episode is that the market did not need to disprove the long-term AI thesis. It only needed to move far enough, quickly enough, to exhaust the portfolio’s tolerance for loss. That is a risk-management failure even if some of the underlying companies eventually recover.

Risk management is often described as a set of mathematical controls—position limits, value-at-risk estimates, stress tests and stop-loss rules. Those tools matter, but a concentrated fund also needs institutional checks that can challenge the founder’s conviction. The relevant questions include who can reduce a position, who can veto additional leverage, how private and public exposures are aggregated, whether short positions truly hedge the same risks, and how quickly lenders can change collateral terms.

A portfolio built around one broad narrative can appear diversified while remaining economically concentrated. Semiconductor manufacturers, data-center operators, power suppliers, networking companies and AI-cloud businesses have different products, but all can depend on the same variables: capital spending, electricity availability, chip supply, credit conditions and confidence in future AI demand. If those variables reverse, the positions can fall together.

Shorting traditional software does not necessarily solve that problem. The hedge assumes that software companies will underperform the AI-infrastructure beneficiaries. During a crowded unwind, investors may buy back the shorts while selling the longs. That produces a loss on both sides. The short book may also have lower volatility, different liquidity or a different sensitivity to rates than the long book. A hedge against the thesis is not automatically a hedge against the financing structure.

Private investments create another layer. Their prices are not continuously observed, which can make the total portfolio look calmer than its public component. Yet they can still require capital calls, have limited liquidity and be exposed to the same AI spending cycle. When the liquid book is sold to meet financing demands, the fund can be left with assets that are harder to monetize precisely when flexibility is most valuable.

A robust stress test would not ask only what happens if AI stocks fall 10%. It would combine several events: a 25% decline in high-beta longs, a 15% rally in software shorts, wider bid-ask spreads, higher margin requirements, delayed private financing and investor redemption requests. The scenario should estimate not merely the mark-to-market loss but also the cash that must be posted within one or two days.

The episode also highlights model risk. Correlations measured during a rising market can understate how quickly assets converge in a selloff. Volatility estimates based on recent calm periods can make leverage appear safer just before it becomes dangerous. A strategy that produced extraordinary gains may even increase confidence in the model at the moment when position size has made the model least reliable.

The governance challenge of a founder-led fund

Situational Awareness was inseparable from Aschenbrenner’s public intellectual identity. He had written an ambitious account of rapid AI progress, national-security competition and the need for enormous computing infrastructure. That clarity helped attract capital. It may also have made dissent more difficult.

Founder-led investment firms can move quickly because one person supplies the thesis, portfolio construction and fundraising narrative. The same concentration of authority can create blind spots. A chief risk officer who cannot constrain the founder is not an independent control. An investment committee composed mainly of believers may debate individual securities while leaving the shared premise untouched. Strong historical returns can make counterparties and investors less willing to ask uncomfortable questions.

None of this proves that internal controls were absent. The public record does not disclose the firm’s governance structure in enough detail to make that judgment. It does show why investors should seek evidence of decision-making discipline rather than infer it from performance. The relevant test is whether controls can force a reduction when the portfolio manager remains convinced.

Who Backed the Fund—and Why the Backers Matter

A fund can grow only as quickly as outside investors and financing counterparties allow. The Wall Street Journal reported on August 5 that Situational Awareness drew backing from experienced figures across technology and finance, including D1 Capital founder Dan Sundheim, Greenoaks co-founder Neil Mehta, a foundation associated with XN founder Gaurav Kapadia and former Tiger Global public-equities chief Feroz Dewan.

Those names change the interpretation of the episode. This was not merely an inexperienced manager finding uninformed capital. Some backers had spent years evaluating technology businesses, hedge funds or both. Their participation suggests that Aschenbrenner’s access, analytical framework and early results persuaded sophisticated people that the opportunity justified unusually high risk.

It also makes the due-diligence questions more consequential. Experienced investors are generally capable of understanding leverage, concentration and liquidity. They may nevertheless accept them when expected returns appear exceptional, when access to a scarce manager is limited or when a portfolio has already produced gains that seem to validate the thesis. Sophistication reduces some forms of misunderstanding; it does not eliminate incentives to chase performance.

The financing counterparties played a separate gatekeeping role. Prime brokers do not need to agree with a fund’s long-term thesis, but their loans and margin terms determine how large the positions can become. Competition for a profitable client can support expansion during good periods. After prices fall, the same institutions protect themselves by raising collateral requirements or reducing exposure. The transition from accommodating lender to risk-reducing lender can be abrupt.

The lesson is not that prominent backers should have predicted the exact July path. It is that reputation cannot substitute for structural analysis. An investor list may validate a manager’s network and fundraising ability, but it says little about whether the portfolio can withstand a correlated shock. The relevant protection comes from enforceable limits, transparent reporting and a financing plan that remains viable when every counterparty becomes cautious at once.

What Institutional Investors Should Have Asked

The reported scale of capital committed to a young manager has prompted easy criticism, but age is not a risk metric by itself. A 24-year-old can be analytically sophisticated, and an older manager can be reckless. The more useful question is what due diligence investors performed before allowing a concentrated strategy to operate with material leverage.

Institutional allocators typically examine the manager, the process, operations and portfolio. In this case, the following questions would have been central:

• How is leverage defined? Investors need gross, net, balance-sheet and derivative-adjusted measures, not one headline ratio.
• Which counterparties can change margin terms? The answer determines whether the fund controls its own investment horizon.
• How much of the portfolio can be liquidated in one, five and 20 trading days? Estimates should account for the fund’s own market impact.
• How concentrated are the true economic drivers? Different tickers may depend on the same chip, power or financing cycle.
• How are private holdings valued? A valuation policy should specify comparable transactions, financing rounds, impairments and independent review.
• Who can override the chief investment officer? A documented answer matters more than an informal assurance.
• What happens after a 20%, 40% or 60% drawdown? The fund should show liquidity, collateral and redemption consequences, not only expected returns.
• Are investor terms aligned? Gates, lockups and side letters can change which investors bear the cost of an unwind.
• How much capital belongs to the manager? Personal investment can align incentives, but it can also intensify commitment to a failing position.
• What evidence would falsify the thesis? A strategy without predefined disconfirming signals can turn every decline into an invitation to add risk.

The spectacular first-half return should have intensified these questions. A 439% gain in six months does not arise from a conventional diversified portfolio. It signals either extreme exposure, extreme luck, an unusually convex payoff or some combination of the three. Investors attracted by the result should have demanded an equally extraordinary explanation of the downside.

Lessons for Individual Investors

Most individual investors will never allocate capital to a private hedge fund, but the mechanisms are familiar. Margin accounts, options, leveraged ETFs and concentrated thematic portfolios can all create a smaller version of the same problem.

1. Separate the thesis from the instrument

An investor can be right that AI infrastructure spending will grow and still choose a vehicle that cannot survive volatility. A leveraged single-stock ETF, a short-dated call option and an unleveraged diversified fund can express a similar direction with radically different paths. The more fragile the instrument, the less time the thesis has to work.

2. Judge leverage by the loss that forces action

The useful question is not “How much can this return if I am right?” It is “What move would force me to sell if I am temporarily wrong?” A position that requires liquidation after a 15% decline is effectively a short-horizon trade, regardless of the investor’s five-year story.

3. Daily reset changes the product

A leveraged ETF generally promises a multiple of one day’s return. Holding it for weeks or months exposes the investor to compounding and volatility drag. The path matters. Two markets that finish at the same index level can produce very different leveraged-ETF outcomes depending on the sequence of daily moves.

4. Diversify economic drivers, not just ticker symbols

Owning a chipmaker, a data-center landlord, a power-equipment company and an AI cloud provider may look diversified by industry label. If all four depend on the same capital-spending boom, the portfolio may behave like one large position during stress.

5. Liquidity disappears when it is most valuable

Trading volume observed in calm markets can overstate how much a position can be sold without affecting price during a shock. Investors should compare position size with normal daily volume, recognize that other holders may sell simultaneously and avoid assuming the last quoted price is available for the entire position.

6. A prior gain does not create a cushion against percentage losses

After a 439% gain, a 67% decline still leaves a positive return from the original starting point, but it destroys most of the accumulated value. Percentage gains and losses are asymmetric. A 67% loss requires a gain of approximately 203% just to recover to the pre-loss level.

7. Private valuations are not the same as liquidity

A private holding may not be marked down every day, but that does not make it low risk. It may be impossible to sell quickly, and the next financing round can reset its value sharply. Apparent stability can be an absence of price discovery rather than an absence of economic change.

8. Reducing leverage after a drawdown is rational but costly

Once a leveraged portfolio has suffered a large loss, deleveraging can protect the remaining capital. It also locks in losses and reduces participation in any rebound. This is why the decision to borrow must be evaluated before volatility arrives. Afterward, every choice can be unattractive.

What Changed After the July Unwind

The most concrete change was the removal of leverage and the sale of most public-equity positions. That transformed Situational Awareness from a highly liquid, actively financed public-market portfolio into a smaller vehicle with a greater relative weight in private investments. It reduced the immediate probability of another margin-driven public-market liquidation, but it also changed the fund’s opportunity set and liquidity profile.

The reported 80% gain for 2026 after the July loss means the fund remained profitable for investors who were present at the beginning of the year, before fees and depending on their exact share class. That fact complicates the “collapse” label. The firm suffered a profound drawdown and lost much of its peak value, yet it did not enter bankruptcy, shut down or report a total loss of investor capital.

For Aschenbrenner, the challenge is now institutional rather than purely analytical. Rebuilding confidence requires more than another correct stock call. Investors will want clearer leverage limits, independent risk controls, liquidity reporting and evidence that future gains will not automatically be recycled into larger exposures. The manager’s willingness to change the structure may matter more than whether the same AI securities rebound.

For the market, the forced sale provided a rare glimpse into the ownership structure of the AI trade. When a concentrated fund becomes large enough, its transactions can affect the prices used to justify its own positions. The reverse is also true: falling prices can expose leverage, prompt selling and reveal that apparent fundamental repricing partly reflects a financing event.

What Happens Next

Several questions remain unresolved as of the August 5 research cutoff.

• Will Situational Awareness raise new capital? The June investor communication reportedly described that period as an attractive time to add funds, but the July drawdown likely changed the terms of any fundraising discussion.
• How will private holdings be valued? Investors will watch whether later financing rounds or transactions alter the marks of retained AI companies.
• Will public positions be rebuilt? Removing leverage does not necessarily mean abandoning the AI thesis. The fund could return to listed securities with smaller sizing and more durable financing.
• Will regulators tighten leveraged-product rules? South Korea has already imposed restrictions, while U.S. authorities continue to emphasize disclosure and suitability. A severe U.S. retail loss event could accelerate policy debate.
• Will semiconductor volatility persist? The answer depends on memory pricing, AI capital expenditure, supply additions, interest rates and the behavior of leveraged vehicles. Flow-based pressure can fade even when fundamental uncertainty remains.
• Will counterparties demand more conservative terms? Prime brokers often revise financing after a visible loss. Higher haircuts and lower concentration limits would reduce return potential but improve survival odds.

The most informative future evidence will not be a single monthly return. It will be the portfolio’s structure: gross and net exposure, the share of liquid assets, concentration by economic driver, collateral requirements and the authority of independent risk personnel. A rebound achieved with the same fragility would not resolve the underlying issue.

Frequently Asked Questions

What happened to Leopold Aschenbrenner’s Situational Awareness fund?

The fund’s portfolio fell 67% in July 2026 after an extraordinary 439% net gain during the first half of the year, according to an investor letter reported by Reuters and The Wall Street Journal. Situational Awareness then removed all leverage and sold most of its public-equity portfolio in a transaction in which Citadel acquired the bulk of the assets. The firm retained a smaller portfolio that included private investments and continued operating.

Did the fund lose $35 billion?

That figure cannot be established by subtracting a reported peak portfolio scale of roughly $45 billion from a later retained-portfolio figure of about $10 billion. The numbers may include different combinations of gross exposure, borrowed capital, derivatives, public securities and private-company valuations. The clearest reported performance figure is the 67% July drawdown, not a verified $35 billion cash loss.

Was the Situational Awareness fund liquidated or closed?

No complete liquidation or closure has been reported. The fund unwound most of its public-stock positions and eliminated leverage, but it retained private investments and continued to exist. Citadel purchased assets; it did not acquire the investment firm itself.

How could a fund fall 67% and still be up about 80% for the year?

Returns compound from the new asset base. A 439% gain turns $100 into $539. A subsequent 67% loss reduces $539 to approximately $178, which is still about 78% above the original $100. Rounding, fees, valuation timing and the exact return figures can explain the difference between that illustration and the approximately 80% year-to-date return reported in the investor letter.

What did Situational Awareness own?

Its March 31 Form 13F disclosed large reportable positions tied to AI infrastructure, semiconductors, memory, data centers and power, including holdings or long options involving Bloom Energy, CoreWeave, Sandisk, Micron, Nvidia, Broadcom, AMD, Oracle and the VanEck Semiconductor ETF. The filing did not show the entire portfolio because it excluded short stock positions, private companies, many foreign securities, written options and numerous swaps.

Was the fund really leveraged four times?

The Prof G Markets discussion described leverage of roughly four times, but the publicly available SEC filing does not establish a precise ratio at the time of the July loss. The investor letter confirmed that leverage was removed after the drawdown, and the forced unwind indicates that financing was material. A definitive multiple would require fuller disclosure of borrowings, derivatives, short positions and net asset value.

Why can both long and short positions lose during an unwind?

A stressed fund may have to sell its long positions while buying back short positions. Those trades push in the wrong direction for both sides of the book: long holdings can fall under selling pressure, while shorted securities can rise as the fund covers. Other market participants may anticipate the orders, magnifying the move.

What is volatility drag in a leveraged ETF?

Volatility drag is the erosion caused by compounding daily percentage changes from a changing base. A daily leveraged ETF resets its exposure each session, so its return over several days can differ substantially from the leverage multiple times the underlying asset’s total return. The effect becomes more severe as leverage and volatility increase.

Did South Korea ban leveraged ETFs?

Not across the board. South Korean authorities limited individual investors’ exposure to single-stock leveraged ETFs, raised minimum-deposit and education requirements, suspended new listings and advertising, and introduced additional stabilization measures after the July market turmoil. Describing those actions as a total ban overstates the verified policy.

Did leveraged ETFs cause the semiconductor selloff?

They may have amplified daily flows, especially in crowded and volatile securities, but the public evidence does not establish them as the sole cause. Semiconductor prices were also responding to valuation, supply expectations, AI capital-spending assumptions, memory-market conditions, macroeconomic uncertainty and forced selling by large investors. Flow mechanics and fundamentals interacted.

Does the drawdown mean the AI investment thesis is wrong?

No. It shows that the portfolio structure was unable to withstand a sharp adverse move. AI infrastructure demand could remain strong while some related securities deliver poor returns because expectations were too high, financing was fragile or supply expanded faster than demand. A valid industry thesis does not guarantee that every security, price or leveraged strategy will succeed.

What should investors watch now?

The most useful indicators are the fund’s future leverage limits, liquidity, concentration, private-asset valuation, counterparty terms and governance. In the broader market, investors should watch semiconductor capital expenditure, memory pricing, leveraged-product assets, prime-broker financing conditions and whether South Korea’s restrictions reduce destabilizing flows without simply shifting them elsewhere.

Final Assessment

Situational Awareness’s July drawdown was not a conventional story of an investor making one bad call. It was a demonstration of how an ambitious macro-technology thesis can become vulnerable when concentrated securities, derivatives, borrowed money and short-term financing are layered on top of one another.

The verified evidence supports several conclusions. The fund generated an exceptional 439% net return in the first half of 2026, lost 67% in July, removed leverage and sold most of its public-equity book. Citadel acquired the bulk of those positions through a transaction facilitated by prime brokers. The fund survived and remained up roughly 80% for the year, but the drawdown erased most of the wealth created during the rally and forced a fundamental restructuring.

The strongest interpretation in Aschenbrenner’s favor is that the long-run AI infrastructure thesis may remain intact. The securities were sold under financing pressure, not necessarily because every underlying business deteriorated. A forced seller and a patient buyer can reach opposite decisions without either taking a different view of technology’s eventual direction.

The strongest concern is harder to dismiss. A strategy that cannot survive an adverse month is governed by its lenders and market path rather than by the time horizon of its research. Extraordinary returns may have concealed the degree to which the fund’s success depended on favorable correlations, abundant financing and the continued willingness of counterparties to support a growing book.

The episode should not be reduced to the age of the manager, a single leverage multiple or a verdict on AI. Its broader significance lies in the interaction between conviction and structure. Investors can be early, insightful and ultimately right about a technological transformation while still losing control of the portfolio built to profit from it.

What matters next is whether Situational Awareness changes the architecture that produced the loss. Smaller positions, less leverage and stronger independent controls would reduce the chance of another forced unwind, though they would also make a repeat of the 439% half-year gain less likely. That trade-off is not a defect in prudent investing. It is the cost of preserving the ability to remain invested long enough for a thesis to be tested by business results rather than by a margin call.

Sources

• Prof G Markets: “Aschenbrenner’s AI Fund Collapse Is Just the Beginning”
• Reuters: Situational Awareness portfolio sinks 67% in July
• Reuters: Citadel buys most of Situational Awareness’s stock holdings
• The Wall Street Journal: Investors who backed Situational Awareness
• The Wall Street Journal: Situational Awareness down 67% in July
• Financial Times: Situational Awareness’s first-half performance and fundraising
• U.S. Securities and Exchange Commission: Situational Awareness LP Form 13F filing
• U.S. Securities and Exchange Commission: Frequently asked questions about Form 13F
• Leopold Aschenbrenner: “Situational Awareness”
• Michael Green: “A Semi-Theory of Almost Everything”
• Roundhill Investments: DRAM ETF fund information
• FINRA: The lowdown on leveraged and inverse exchange-traded products
• Direxion: Why volatility matters for daily leveraged ETFs
• Federal Reserve: May 2026 Financial Stability Report—leverage in the financial sector
• Office of Financial Research: Calm markets and underlying risks
• Reuters: South Korea limits investment in single-stock leveraged ETFs
• U.S. Securities and Exchange Commission: Investor bulletin on margin accounts

This article is provided for general informational purposes and does not constitute financial, investment, tax, or legal advice.

Affiliate disclosure: Businessfinance.news may earn compensation from qualifying actions completed through selected links on this website, at no additional cost to the reader. Affiliate relationships do not influence our editorial reporting, analysis, or conclusions.