How Situational Awareness Lost 67% on the AI Trade

Last updated: August 6, 2026, 4:00 a.m. EDT

Situational Awareness, the fast-growing hedge fund founded by former OpenAI researcher Leopold Aschenbrenner, lost about 67% in July 2026 after a violent reversal in artificial-intelligence-linked stocks triggered margin calls and forced the firm to sell most of its public-equity portfolio to Citadel. The sale covered roughly $16 billion of public holdings and was completed rapidly enough to prevent a disorderly liquidation from spilling more heavily into the market. The fund’s reported year-to-date return remained positive—about 80%—because its gains earlier in 2026 had been extraordinary, but the July drawdown represented a severe destruction of capital and a near-loss of control over the portfolio.

The episode is often summarized as a $45 billion hedge-fund collapse. That shorthand is dramatic but imprecise. The available reporting indicates that Situational Awareness had reached as much as roughly $45 billion in assets or portfolio exposure at its peak after combining investor capital, gains, private-company marks and substantial leverage. It does not mean that outside investors handed the firm $45 billion in cash, nor does it establish that exactly $45 billion was lost. Public filings show a much smaller—but still unusually large—snapshot: the firm’s Form 13F for March 31, 2026 listed $13.68 billion of reportable U.S. securities and option-related underlying exposure. Form 13F is not a balance sheet, excludes many asset types and does not disclose borrowing, so it cannot be used by itself to calculate the fund’s net asset value or true leverage.

What is clear is more important than the headline arithmetic. A concentrated investment thesis about AI infrastructure produced enormous early profits. Those profits attracted fresh capital and encouraged larger positions. Borrowing amplified the exposure. When several crowded holdings fell together, the firm’s lenders demanded additional collateral. The resulting need for immediate liquidity overrode the manager’s long-term conviction. The portfolio had to be sold on someone else’s timetable and at a reported discount of more than 10%.

That sequence turns Situational Awareness into a useful case study in modern market risk. The underlying AI thesis may still prove broadly correct. Demand for data centers, semiconductors, power equipment, memory, networking and cloud capacity remains substantial. Yet a correct long-term thesis does not protect a leveraged fund from a short-term funding crisis. Markets can force a sale before the thesis has time to mature. In that sense, the central lesson is not that artificial intelligence is a fraud or that every AI stock is in a bubble. It is that leverage converts uncertainty about timing into a threat to survival.

Key Takeaways

• Main development: Situational Awareness lost approximately 67% in July 2026 and sold most of its public-equity portfolio to Citadel after margin calls.
• Peak scale: Reporting placed the fund’s peak portfolio or asset figure near $45 billion, but that number included leverage and should not be confused with investor capital or a precisely measured cash loss.
• Public filing evidence: The firm’s March 31, 2026 Form 13F reported $13.68 billion across 42 entries, including large put and call positions stated in terms of underlying securities.
• Portfolio design: The disclosed book paired long positions in AI infrastructure beneficiaries with large put positions in semiconductor and technology names, making it more complex than a simple long-only AI fund.
• Why it failed: Concentration, leverage, correlated price moves, financing terms and the limited liquidity of some holdings combined to create a margin-call spiral.
• Why it was not another LTCM: The positions were large enough to affect individual stocks, but the sale to Citadel appears to have contained the immediate market impact rather than creating a systemwide rescue.
• Why the AI thesis survives: Alphabet, Amazon, Meta and Microsoft continue to spend heavily on AI infrastructure, while cloud and AI revenue are growing. The debate is increasingly about returns on capital, pricing and timing—not whether spending exists.
• What comes next: Investors will focus on the fund’s remaining private holdings, its financing relationships, its next investor communication and later regulatory filings. The next 13F will reflect June 30 positions and therefore will not fully capture the late-July sale.

What Happened to the Situational Awareness Hedge Fund?

The immediate answer is a classic leveraged-fund failure compressed into a remarkably short period. Situational Awareness entered July with a portfolio that had appreciated dramatically and with significant exposure to AI-related public companies. The trade had become large, concentrated and dependent on borrowed money. A sharp selloff in several core holdings reduced the value of collateral supporting the loans. Prime brokers and other lenders demanded more cash or lower exposure. The fund could not satisfy those demands solely by waiting for prices to recover, because margin calls are contractual and time-sensitive. It therefore sought a buyer for a large block of positions.

Citadel acquired the majority of the public-equity book in a transaction that reporting placed at around $16 billion. The positions were reportedly sold at a discount exceeding 10%, reflecting both their size and the seller’s limited bargaining power. Other firms, including Millennium Management, were said to have considered the portfolio. A single-buyer transaction reduced execution complexity and the risk that multiple buyers would selectively take only the most liquid or attractive assets while leaving the seller with the hardest positions to exit.

The difference between an orderly block transfer and open-market liquidation matters. If a distressed fund sells billions of dollars of overlapping positions through the market, prices can fall further simply because the seller must keep selling. Those declines can trigger additional margin calls, forcing more sales. Other investors holding the same names may then reduce risk, and short sellers may press the weakness. A negotiated portfolio transfer can interrupt that loop by moving the assets to a buyer with enough capital and time to hold them.

Citadel appears to have benefited quickly. Reuters reported that Citadel’s Wellington fund gained about 6% in July, while its tactical-trading and equities funds rose roughly 11% and 14.2%, respectively. The Wall Street Journal and Financial Times linked part of those returns to the discounted Situational Awareness purchase. The gains do not prove that every acquired position immediately recovered or that the ultimate transaction profit is known. They do show the asymmetry between a forced seller and a well-capitalized buyer: the buyer can demand a discount for assuming market, liquidity and execution risk at the moment the seller has the least flexibility.

Situational Awareness did not disappear. Reporting indicated that it retained private investments, including a large stake in Anthropic, as well as interests in other AI-related companies. The firm therefore shifted from a highly liquid—or at least publicly tradable—portfolio toward a more private and valuation-dependent asset base. That can reduce daily market volatility, but it creates different risks. Private holdings cannot necessarily be sold quickly, their marks may lag public-market conditions, and a valuation is not the same as cash available to meet redemptions or margin calls.

In an investor letter described by Reuters and The Wall Street Journal, Aschenbrenner acknowledged that the firm had come closer to permanent capital impairment than it considered acceptable and told investors, “We let you down this month.” That admission is notable because it frames the problem as more than a bad month. A 67% drawdown requires a gain of slightly more than 203% merely to return to the starting value. Even when a fund remains positive for the year, the path matters: large losses can permanently alter investor confidence, financing terms, employee retention and the manager’s willingness to take risk.

From a 2024 AI Manifesto to a Wall Street Sensation

Situational Awareness was unusual from the beginning because its investment identity was inseparable from an intellectual argument. In June 2024, Aschenbrenner published a long series of essays titled Situational Awareness: The Decade Ahead. The central thesis was that improvements in computing scale and algorithmic efficiency could produce artificial general intelligence within a few years, followed by an even faster transition toward superintelligence. The essays argued that the economic and national-security consequences would be immense and that the United States would need to mobilize industrial capacity, power infrastructure and security resources on a scale rarely seen outside wartime.

The investment implication was straightforward even if the forecast was not. If frontier AI systems required rapidly expanding compute clusters, then the most valuable bottlenecks would include advanced chips, high-bandwidth memory, data-center capacity, electrical generation, cooling, networking and specialized infrastructure. Companies controlling scarce capacity could experience revenue and pricing growth well before the full economic benefits of AI appeared in end-user applications. A fund built around that proposition could buy the “picks and shovels” of the AI buildout rather than trying to predict which chatbot or software product would eventually dominate.

Aschenbrenner had relevant technological credibility but little conventional portfolio-management experience. He had worked on OpenAI’s Superalignment team and contributed to research on weak-to-strong generalization. His own biography identifies early backing for the investment firm from Patrick Collison, John Collison, Nat Friedman and Daniel Gross. The fund subsequently attracted other wealthy technology founders, investors and finance professionals. That network mattered because new hedge funds normally struggle to raise capital without a long audited track record, an established risk organization and years of institutional references.

Situational Awareness’s rise was therefore partly an investment story and partly a social-capital story. Aschenbrenner had a clear narrative, access to influential technology circles and a thesis that appeared to be validated by market prices. Early success made later fundraising easier. Later fundraising allowed the firm to increase positions. Increasing positions amplified the performance of the original thesis. This is a familiar reflexive process: returns attract assets, assets increase market influence, and market influence can reinforce returns until the direction reverses.

Public reporting traces a dramatic acceleration. The Wall Street Journal reported in August 2025 that the firm managed more than $1.5 billion. By June 2026, another Journal report placed assets above $20 billion after exceptional performance and inflows. Financial Times reporting later said the fund had gained about 439% in 2026 before the July collapse and had reached roughly $24 billion in assets, while other accounts described a peak figure near $45 billion when leverage and broader portfolio exposure were included. The varying numbers are not necessarily contradictory; they may refer to different dates, valuation methods, legal entities, gross exposure, net assets or private-company marks.

The speed was extraordinary. Traditional hedge funds generally build capacity gradually because adding capital can dilute returns, strain operations and increase market impact. A concentrated strategy has an additional problem: the manager can run out of liquid securities that fit the thesis. The temptation is then to enlarge existing positions, use derivatives, expand into smaller companies or add leverage. Each choice makes the portfolio more sensitive to liquidity and correlation.

The fund’s origin also shaped its risk culture. A technology researcher approaching markets from first principles may see securities as claims on a powerful underlying industrial trend. A veteran trader is more likely to focus first on path dependency: how much a position can move before the thesis is proved, who else owns it, how it is financed, what collateral lenders will accept and what happens if several positions become correlated during stress. Both perspectives can be valuable. The failure occurred because the second set of questions became decisive before the first set could be resolved.

Timeline: The Rise, the Leverage and the Forced Sale

1. June 2024: Aschenbrenner published Situational Awareness: The Decade Ahead, setting out an aggressive forecast for AI capability growth and the infrastructure required to support it.
2. 2024: Situational Awareness Partners LP and related entities were formed. A September 2024 Form D filing documented the private fund structure.
3. March 31, 2025: The first public 13F snapshot showed a comparatively small but concentrated U.S. portfolio, including Applied Digital, Broadcom, Core Scientific, CoreWeave, Intel call options and power producers.
4. August 2025: The Wall Street Journal reported that the fund had grown beyond $1.5 billion and was part of a broader wave of AI-focused hedge-fund launches.
5. Late 2025: Regulatory filings showed growing positions and beneficial ownership in selected AI-infrastructure companies. The strategy expanded beyond large technology companies into data centers, power, bitcoin miners converting capacity to AI use and specialized hardware.
6. March 31, 2026: The fund’s 13F reported $13.68 billion across 42 entries. Large long positions included Bloom Energy, Sandisk, CoreWeave, IREN, Core Scientific and Applied Digital. Large reported puts referenced the VanEck Semiconductor ETF, Nvidia, Oracle, Broadcom, AMD, Micron, TSMC, ASML and Intel.
7. May–June 2026: Public reporting described gains of about 270% through May and more than 400% by the end of June, while assets and gross exposure expanded rapidly.
8. July 2026: AI-related equities reversed sharply. Several core holdings fell together, reducing collateral values and triggering margin calls.
9. July 29–30, 2026: Situational Awareness negotiated with potential buyers and sold most of its roughly $16 billion public-equity portfolio to Citadel in a transaction reportedly completed within about 24 hours.
10. July 31, 2026: Reuters reported that the fund’s portfolio had lost approximately 67% in July but remained up about 80% for the year.
11. August 5–6, 2026: Additional reporting showed that Citadel’s July performance had benefited materially from the discounted purchase, while investors and advisers debated whether the episode represented an isolated failure or a warning about broader AI-trade leverage.

The timeline illustrates why point-in-time filings can be misleading. The March 31 portfolio was reported in mid-May. By the time readers saw it, the fund had already had weeks to change positions. By July, both the scale and financing of the book had evolved further. The next 13F, due in mid-August for positions held on June 30, will still predate the late-July sale. A public investor may therefore wait until the filing for September 30—normally due in mid-November—to see a regulatory snapshot that reflects the transaction, and even that filing will omit private assets, many derivatives, borrowing and cash.

What the $45 Billion Figure Actually Means

The most widely repeated number in the story is $45 billion. It is also the number most likely to be misunderstood. In asset management, “assets,” “assets under management,” “gross assets,” “gross exposure,” “net exposure,” “portfolio value” and “investor capital” are different measures. A leveraged long-short fund can report gross exposure far above its net asset value because it owns long positions, holds short positions and uses derivatives that create additional economic exposure.

Consider a simplified example. A fund with $10 billion of investor capital might borrow $20 billion and hold $30 billion of long positions. It could also short $10 billion of securities. Its gross exposure would be $40 billion—the absolute value of $30 billion long plus $10 billion short—while net market exposure would be $20 billion long. Its net asset value would still begin near $10 billion, before fees and market changes. A headline that calls it a “$40 billion fund” could be referring to gross exposure rather than capital entrusted by investors.

Options complicate the picture further. Under the SEC’s Form 13F instructions, managers report option entries in terms of the securities underlying the options, not the market value of the option premium. That means the $13.68 billion total in Situational Awareness’s March filing should not be interpreted as $13.68 billion of cash paid for securities or even as a direct measure of economic delta. About $8.46 billion of the filing’s stated value related to put-option underlying securities, approximately $1.36 billion related to call-option underlying securities and about $3.86 billion related to common shares. The option entries can therefore make the headline filing value much larger than the actual cash committed.

Nor can the filing reveal whether a put position was a directional bearish bet, a hedge against a correlated long, part of a spread, or an offset to exposure held outside the filing. The March snapshot showed enormous put-related entries against semiconductor and technology companies while the fund was widely described as bullish on AI. That apparent contradiction disappears if the strategy was attempting to distinguish between winners and losers within the AI supply chain, hedge market beta or protect a long book. Without strike prices, expiration dates, premiums, counterparties and the rest of the portfolio, the exact risk cannot be reconstructed.

The reported $45 billion peak should therefore be treated as an approximate scale indicator. It tells readers that the fund had become extremely large relative to its age and team. It does not provide a clean denominator for calculating losses. The reported decline to around $10 billion likewise may combine remaining private holdings, cash, net asset value and updated marks in a way that is not publicly auditable. The more defensible statement is that a highly leveraged portfolio with a reported peak scale near $45 billion suffered a 67% monthly loss and transferred roughly $16 billion of public positions to Citadel.

Inside the March 2026 Portfolio: Long Bottlenecks, Shorter-Lived Hedges

The March 31 Form 13F provides the clearest public window into the fund before the collapse. It listed 42 entries with a total stated value of $13.68 billion. The composition was not a conventional collection of large-cap technology stocks. The long common-stock book was concentrated in companies positioned around energy generation, data-center construction, high-performance computing and memory. The options book, meanwhile, included large puts on many of the semiconductor companies normally associated with the AI boom.

The largest disclosed common-stock position was Bloom Energy, valued at approximately $879 million at quarter-end. Bloom develops fuel-cell systems that can provide on-site power, an attractive proposition for data centers facing delayed grid connections and rising electricity demand. Sandisk was next at roughly $724 million, reflecting the importance of storage and memory economics. CoreWeave, a specialized cloud provider built around GPU capacity, represented about $556 million in common shares plus call-option underlying exposure stated at roughly $141 million. IREN, Core Scientific and Applied Digital—companies with origins in bitcoin mining or specialized data-center infrastructure—together accounted for more than $1.1 billion of disclosed common-stock value.

That grouping reveals the fund’s differentiated thesis. Rather than simply buying Nvidia and the largest cloud platforms, Situational Awareness appeared to favor second-order beneficiaries where capacity constraints could produce sharper earnings changes. A power supplier or data-center developer can experience nonlinear upside if scarce megawatts become more valuable. A memory producer can benefit when high-bandwidth memory supply is tight. A former crypto miner can re-rate if power contracts and land are repurposed for AI workloads.

The same characteristics also create vulnerability. Smaller infrastructure companies often have higher financing needs, less diversified revenue and more volatile shares than megacap technology firms. Their valuations may depend on future contracts, construction schedules, power availability and capital-market access. A stock that rises several hundred percent on expectations can fall quickly when investors question project timing, customer concentration or the cost of expansion. If a hedge fund owns a large percentage of the tradable float, exiting can be difficult without moving the price.

The put side of the filing was even larger in stated underlying value. The VanEck Semiconductor ETF accounted for approximately $2.04 billion. Nvidia puts represented about $1.57 billion; Oracle, $1.07 billion; Broadcom, $1.01 billion; AMD, $969 million; Micron, $584 million; TSMC, $535 million; ASML, $494 million; and Intel, $159 million. In aggregate, put-related underlying value was about 61.9% of the filing total.

Those numbers should not be read as proof that the fund was net short the semiconductor industry. Form 13F does not report the premium paid, strike, maturity or option delta. A far-out-of-the-money put can reference a large amount of underlying stock while contributing relatively little day-to-day sensitivity. Puts can also hedge long positions in suppliers whose shares tend to move with the semiconductor index. The filing nevertheless shows a portfolio with substantial derivative complexity, not a simple cash-equity book.

That complexity matters in a stress event. Options can protect against a broad market decline, but hedges do not always offset the exact losses in a concentrated portfolio. A put on Nvidia or a semiconductor ETF may gain when chip stocks fall, yet it may not compensate for a much larger collapse in a data-center developer, power company or memory supplier. Correlations that appeared stable in ordinary markets can change abruptly. Implied volatility can rise, bid-ask spreads can widen and the timing of hedge gains may not match the timing of collateral demands on the long book.

How Leverage Turned a Drawdown Into a Survival Event

Leverage is often described as a return enhancer, but that definition is incomplete. Leverage changes who controls the holding period. An unleveraged investor can decide to tolerate a 30% decline if the thesis remains intact. A leveraged investor may not have that choice. The lender can demand more collateral, raise financing costs, reduce allowable exposure or liquidate positions under the terms of the agreement.

The basic arithmetic is unforgiving. Suppose a fund begins with $10 billion of equity and borrows $30 billion to hold $40 billion of assets. Its gross asset-to-equity ratio is four to one. A 10% decline in the assets erases $4 billion, reducing equity to $6 billion before fees and financing costs—a 40% loss to investors. The leverage ratio then rises automatically because the asset base has not fallen as quickly as the equity cushion. If the lender requires the fund to restore the original ratio, the manager must sell assets into weakness or inject new capital.

A 20% asset decline in the same example would erase $8 billion, leaving only $2 billion of equity. The fund might still own valuable assets, but its creditors would be exposed to a much thinner cushion. They would have strong incentives to demand repayment before prices fell further. The manager’s conviction about long-term AI demand would be irrelevant to the immediate financing decision.

Reported leverage near four times would therefore be enough to explain how a severe but not unprecedented sector reversal produced a catastrophic fund-level loss. It also helps explain why the portfolio could grow from a much smaller capital base to a headline scale near $45 billion. The same mechanism that magnified early gains magnified the reversal.

Leverage can arrive through several channels. A prime broker may lend cash against equities. Total-return swaps can provide synthetic exposure without the fund owning the shares directly. Options can create nonlinear exposure for a relatively small premium. Repurchase agreements can finance securities. A fund may also borrow at the portfolio-company level or invest in businesses that themselves carry substantial debt. Public filings do not provide a complete map of these layers.

The Office of Financial Research warns that leveraged hedge funds depend on creditors’ willingness to continue lending. Declining collateral values can produce margin calls that consume liquid assets and impair a fund’s ability to meet short-term funding needs. The Federal Reserve’s May 2026 Financial Stability Report said hedge-fund leverage remained high and concentrated among the largest funds. Its July Monetary Policy Report went further, saying leverage was near all-time highs. Situational Awareness therefore failed against a backdrop in which leverage was already a recognized market vulnerability, not an isolated technical detail.

There is a second-order effect. A lender does not need to believe the underlying securities are worthless. It only needs to believe that the borrower’s equity cushion is shrinking too quickly. Prime brokers also react to concentration, volatility, liquidity and the behavior of other lenders. If one bank raises margin, others may follow to avoid being the last creditor left with weak collateral. That coordination problem can accelerate a fund’s collapse even when no single lender wants a fire sale.

Risk limits are supposed to prevent this outcome. They can include maximum gross exposure, maximum position size, minimum liquidity, stress-loss thresholds, counterparty diversification and pre-agreed reduction plans. Yet limits often loosen after strong performance. A position that began as 5% of capital can become 20% because the stock rises. A manager may view that growth as validation and resist trimming. New inflows can then be allocated to the same winners, preserving the concentration. If volatility remains low, risk models may show the position as safer precisely when its market crowding and valuation risk are increasing.

The Margin-Call Spiral, Step by Step

A margin call is not a judgment about a manager’s intelligence. It is a funding event. Understanding the sequence helps explain why a fund can move from celebrated performance to forced liquidation in days.

1. The collateral falls

The value of financed positions declines. In a concentrated portfolio, several securities may fall together because they share the same factor exposure. For Situational Awareness, that factor was not simply “technology.” It included expectations for AI infrastructure spending, access to financing, power scarcity, memory pricing and investor willingness to pay high multiples for future capacity.

2. Volatility increases the lender’s required cushion

Prime brokers calculate how much collateral they need based partly on how quickly a position can move and how difficult it would be to liquidate. When volatility rises, a loan that was acceptable at a 15% haircut may require a 25% or 35% haircut. The borrower can face a margin call even if the position’s price has not fallen further, because the financing terms have tightened.

3. Concentrated positions receive harsher treatment

A lender may finance a diversified S&P 500 portfolio more generously than a large block in a volatile mid-cap company. If the fund owns a meaningful portion of average daily trading volume, the broker cannot assume it can exit at the screen price. It models a liquidation discount. The larger and less liquid the position, the more collateral the broker may demand.

4. The fund sells its most liquid assets first

Managers under pressure often sell what they can rather than what they would prefer. That can leave the remaining portfolio less liquid and more concentrated. Cash-generating hedges may be closed to meet immediate obligations, reducing protection against further declines. A fund can therefore become riskier even as it shrinks.

5. The sales push prices down

If market participants identify a forced seller, they have little incentive to provide generous prices. Buyers wait, reduce bids or demand block discounts. The fund’s own liquidation then worsens the marks used by its lenders, producing another round of calls.

6. Time becomes the scarcest asset

At this point, the manager needs a capital injection, a portfolio buyer or concessions from lenders. New investors know the fund is under pressure. Existing investors may be unable or unwilling to add capital. Lenders may extend time only if they believe a credible transaction is near. The bargaining advantage shifts away from the fund.

Situational Awareness’s sale to Citadel appears to have interrupted the process at the final stage. Rather than continue selling names individually, the firm transferred a large portfolio to a buyer capable of absorbing both the market exposure and the operational complexity. That reduced immediate liquidation pressure but crystallized a discount and surrendered future upside on the sold assets.

Why Citadel Was the Natural Buyer

Citadel’s advantage was not merely that it had cash. Large multistrategy firms combine capital, financing relationships, execution technology, sector specialists, risk systems and the ability to distribute positions across multiple internal teams. A portfolio that is too large or dangerous for one concentrated fund can be manageable when broken into smaller risk buckets inside a much larger organization.

The transaction also fit a recurring pattern in Ken Griffin’s career. Citadel has previously acquired or financed portfolios during periods of distress, including positions associated with Amaranth Advisors and Sowood Capital. Distressed transfers can be attractive because the seller pays for immediacy. The buyer receives a discount, information about the portfolio and the option to hedge or unwind positions gradually.

For Citadel, the key question was not whether every Situational Awareness investment was good. It was whether the portfolio could be bought below its risk-adjusted liquidation value. A 10% discount on a $16 billion portfolio represents roughly $1.6 billion of immediate price concession before hedging costs, financing, market moves and any differences between the reported figure and the actual transferred assets. Even a partial recovery can create large profits if the buyer manages the exposure efficiently.

Reuters reported that Citadel’s equities fund gained 14.2% in July, its tactical-trading fund rose about 11% and Wellington gained roughly 6%. The Wall Street Journal reported that Wellington had been up only about 0.45% through July 24 before finishing the month near 5.9%, suggesting that the transaction materially affected performance. Those figures are fund-level returns, not a precise profit statement for the acquisition. They nevertheless indicate that Citadel converted another firm’s liquidity crisis into a profitable opportunity.

The deal may also have benefited the broader market. The Financial Times reported that the Nasdaq 100 and several affected stocks stabilized after the transfer reduced fears of continued forced selling. That does not make Citadel a public rescuer in the same sense as the consortium that recapitalized Long-Term Capital Management. Citadel acted for its investors and negotiated a favorable price. The stabilizing effect was a consequence of the transaction, not its charitable purpose.

Situational Awareness’s decision to use one buyer had trade-offs. Selling positions separately might have produced better prices for the most attractive assets, but it would have taken longer and exposed the fund’s distress to more counterparties. A package sale can preserve confidentiality, reduce basis risk and give lenders confidence that the problem will be resolved. The cost is that the buyer prices the weak and strong positions together and demands compensation for assets it does not especially want.

Why This Was Not Long-Term Capital Management 2.0

Michael Novogratz and Anthony Scaramucci compared the event with Long-Term Capital Management, the celebrated quantitative hedge fund that nearly failed in 1998. The comparison is useful, but only if the differences are kept in view.

LTCM had approximately $4.8 billion of equity at the start of 1998 and lost more than 90% of its capital by September. Its balance-sheet positions exceeded $100 billion, while its derivatives had notional value above $1 trillion. The fund traded across government bonds, swaps, options and relative-value positions linked to many global markets. Regulators and major dealers feared that an uncontrolled liquidation would amplify already severe stress following Russia’s default and disrupt counterparties around the world. The Federal Reserve Bank of New York facilitated discussions that led private financial institutions to provide a $3.625 billion recapitalization.

Situational Awareness was large and leveraged, but its visible public book was concentrated in equities and equity derivatives tied to AI infrastructure. Its losses affected specific technology and Korean-market names and created meaningful block-trade risk. Available reporting does not indicate that its failure threatened the solvency of major banks or the functioning of core funding markets. Citadel could buy the portfolio without a regulator-organized consortium.

The common element is the interaction of leverage and liquidity. LTCM believed its convergence trades were fundamentally sound, but the positions became correlated during a crisis and could not be exited at modeled prices. Situational Awareness believed its AI infrastructure thesis was fundamentally sound, but correlated declines and margin requirements shortened the available time horizon. In both cases, the manager’s expected long-term value became subordinate to the market’s immediate liquidation value.

The systemic distinction matters. A fund can suffer one of the largest percentage losses in modern hedge-fund history without creating a financial crisis. Systemic risk depends on interconnectedness, counterparty exposure, substitutability, market depth and the importance of the affected assets—not only the dollar amount lost. The Situational Awareness episode was a severe private failure with localized market consequences. LTCM was a potential transmission mechanism across the global financial system.

The Closer Comparison: Archegos and the Hidden Cost of Concentration

Archegos Capital Management offers a closer modern parallel. In March 2021, the family office run by Bill Hwang failed after concentrated equity positions financed through total-return swaps moved sharply against it. Multiple prime brokers had extended leverage without a complete view of the exposure held at competing banks. When Archegos could not meet margin calls, banks liquidated large blocks of shares. Credit Suisse ultimately recorded a net charge of about CHF 4.8 billion related to the episode.

The resemblance lies in concentrated equity exposure, reliance on prime-broker financing and the speed with which collateral pressure overwhelmed the manager. Both cases also demonstrate that apparently sophisticated counterparties can collectively provide more leverage than they would choose if each had a complete picture of the borrower’s aggregate risk.

There are important differences. Archegos used swaps that concealed much of its beneficial exposure from ordinary public ownership disclosures. Situational Awareness filed 13F reports and beneficial-ownership forms for some holdings, giving the market partial visibility. Archegos also inflicted large direct losses on several banks because the collateral proved insufficient when the positions were liquidated. Public reporting on Situational Awareness has not identified comparable losses at its prime brokers. The Citadel sale appears to have allowed lenders to reduce exposure before a disorderly default.

The comparison nevertheless raises a due-diligence question: how did a young firm with a small investment team obtain enough financing to build such a large and concentrated portfolio? The answer may be that the fund had produced spectacular returns, held valuable private assets, attracted credible backers and traded securities that lenders initially considered liquid. Each factor could justify credit on its own. Combined, they may have created excessive confidence.

Prime brokers compete for profitable hedge-fund clients. A fast-growing fund generates trading commissions, financing revenue, stock-loan fees and derivatives business. If one bank imposes conservative terms, the client can move activity to another. That competitive pressure does not eliminate risk controls, but it can weaken the discipline that lenders are expected to provide. The Federal Reserve has repeatedly emphasized that creditor and counterparty discipline is the primary check on leverage at private funds. Situational Awareness suggests that discipline can arrive late—after the borrower’s rapid success has already encouraged expansion.

The Strongest Case for the AI Infrastructure Thesis

The fund’s failure should not be confused with proof that the AI infrastructure boom is imaginary. The strongest bullish evidence is visible in the financial statements of the world’s largest technology companies. They are committing hundreds of billions of dollars to servers, chips, networking, data centers, land and power because demand for AI and cloud workloads continues to exceed available capacity in important areas.

Alphabet reported capital expenditure of $44.9 billion in the second quarter of 2026, with the vast majority directed to technical infrastructure supporting AI. About 60% of that technical-infrastructure investment went to servers and 40% to data centers and networking. The spending pushed quarterly free cash flow to negative $5.9 billion, yet Alphabet still held substantial cash and marketable securities and generated $39.1 billion of operating cash flow during the quarter. That is an important distinction: the company is spending aggressively from a position of strong underlying cash generation.

Microsoft spent $41 billion in capital expenditures in its fiscal fourth quarter ended June 30, 2026. Roughly two-thirds was directed to shorter-lived assets, primarily CPUs and GPUs, while the rest went to longer-lived assets such as data-center sites. Microsoft’s quarterly operating cash flow reached $55.4 billion and revenue rose 18% to $90 billion. The company therefore offers evidence that AI infrastructure demand and monetization can grow together, even though the spending reduces near-term free cash flow and increases depreciation.

Amazon said in February that it expected about $200 billion of capital expenditures across the company in 2026. In the second quarter, AWS revenue rose about 37% to $42.2 billion and AWS operating income increased to $16.6 billion. Amazon’s trailing-12-month free cash flow turned negative by $7.6 billion, largely because property-and-equipment purchases increased by $66.1 billion year over year, primarily for AI. The cloud business is producing strong operating income, but the infrastructure build is consuming cash faster than depreciation appears in reported earnings.

Meta narrowed its 2026 capital-expenditure outlook to $130 billion to $145 billion after spending $31.08 billion in the second quarter. Revenue grew 28% to $60.8 billion, but free cash flow fell to $784 million from $8.55 billion a year earlier as spending, legal charges and other costs increased. Meta’s position is different from the cloud providers because it has historically monetized AI through better advertising, recommendations and engagement rather than by selling large volumes of compute. That makes the return on infrastructure harder for outside investors to isolate.

These figures support several parts of the Situational Awareness thesis. Compute demand is real. The largest buyers have the balance sheets to continue spending. Power, memory and data-center capacity remain strategic bottlenecks. AI-related cloud revenue is growing rapidly. Companies that can deliver scarce infrastructure may earn attractive returns.

The thesis also benefits from the breadth of use cases. AI workloads include training frontier models, serving inference requests, coding assistants, advertising systems, search, enterprise software, robotics, autonomous vehicles, scientific research and cybersecurity. Even if one application disappoints, aggregate demand can remain strong. Efficiency improvements can lower the cost per task and stimulate more usage, a pattern often described as Jevons’ paradox.

Finally, the fund’s remaining private exposure could still be valuable. Anthropic has become one of the leading frontier-model developers, and strategic investments by Amazon and others have created large accounting gains as private valuations increased. A substantial stake could recover a meaningful portion of Situational Awareness’s losses if Anthropic continues to grow, raises capital at higher valuations or eventually completes an initial public offering. That upside is real, but it is contingent and illiquid.

The Strongest Skeptical Case: Good Technology Can Still Produce Bad Investments

The skeptical interpretation begins with a distinction that financial markets repeatedly forget: an industry can transform the economy while many of its investors lose money. Railroads, fiber-optic networks, airlines, personal computers and the internet all created enormous social value. They also produced bankruptcies, overcapacity and long periods of poor returns for capital providers.

AI infrastructure is becoming more capital-intensive at the same time that the underlying technology is improving rapidly. That creates a vintage risk. A data center built around today’s chips, cooling systems and model architecture may face lower utilization or pricing if newer hardware delivers substantially more inference per watt. A facility can remain physically useful while earning less than expected. The owner then bears depreciation, interest and power-contract obligations that were negotiated under more optimistic assumptions.

Token pricing presents another challenge. The cost of generating AI output has fallen rapidly as models, chips and software improve. Lower prices can increase demand, but they can also compress provider margins. If multiple frontier-model companies and open-weight alternatives offer similar capabilities, customers may switch providers or negotiate discounts. Infrastructure built on expectations of persistent premium pricing could underperform even while token volumes rise.

Capital expenditure is also growing faster than free cash flow at several major companies. Alphabet’s negative quarterly free cash flow, Amazon’s negative trailing-12-month free cash flow and Meta’s sharp free-cash-flow decline are not evidence of insolvency; each company has significant resources. They are evidence that the buildout has an opportunity cost. Cash spent on servers and data centers cannot simultaneously be used for buybacks, acquisitions, dividends or other projects. Investors eventually require proof that incremental AI revenue and cost savings justify the incremental capital.

The economics become more fragile outside the hyperscalers. Specialized cloud providers and data-center developers may rely on a small number of customers, expensive debt, equipment leases and long construction schedules. Their revenue can be contractually strong while their cash flow remains weak. If a customer delays deployment, renegotiates a contract or encounters its own financing problem, the infrastructure owner may have few alternatives. Power interconnection and permitting delays can add another mismatch between spending and revenue.

Private-company valuations can amplify optimism. A fund holding Anthropic shares may report a large gain when a new funding round establishes a higher valuation. That mark improves reported net asset value, but it does not produce cash unless shares are sold. The round may involve preferences, strategic terms or a small portion of the company that are not fully comparable with the fund’s holdings. If the broader AI market weakens, the mark can remain stale until another transaction occurs.

The final skeptical point is portfolio construction. Even if every company in the AI supply chain eventually grows, investors can overpay. A stock priced for years of uninterrupted execution can fall sharply after a merely good quarter. Leverage makes that valuation sensitivity lethal. Situational Awareness’s loss therefore does not invalidate its industry research; it shows that the portfolio was not robust to uncertainty in timing, price and financing.

Why Stops Alone Would Not Have Solved the Problem

Scaramucci and Novogratz framed the risk-management lesson in memorable terms: a trader may use leverage with tight stops, or avoid stops with little leverage, but combining high leverage with no effective exit discipline is dangerous. The principle is sound, yet institutional portfolios require a more detailed version.

A stop-loss order is an instruction to sell after a price reaches a specified level. It can prevent a small loss from becoming a larger one in liquid markets. It cannot guarantee the execution price during a gap, trading halt or fast market. It can also be counterproductive in volatile securities, forcing a sale during a temporary move and leaving the manager unable to participate in a recovery.

For a large fund, the relevant “stop” is often a risk budget rather than a literal order. The manager may reduce exposure when a position exceeds a volatility target, when the portfolio drawdown reaches a limit, when liquidity deteriorates or when financing terms change. Those rules must be established before the crisis. Once a position becomes too large to sell without moving the market, the fund no longer has a clean stop. It has a negotiation.

A robust system would examine at least five dimensions. First is position size: how much can be lost if the security falls 20%, 40% or 70%? Second is liquidity: how many days would it take to sell under normal and stressed volume? Third is factor concentration: how much of the portfolio depends on the same AI-capex narrative even if the tickers are different? Fourth is funding: how much additional collateral could lenders demand if volatility doubles? Fifth is path dependency: can the fund survive long enough for the thesis to work?

Traditional value-at-risk models may understate these dangers because they rely on historical correlations and relatively short return windows. A portfolio can appear diversified across power, memory, cloud and data centers while all positions are exposed to the same shift in investor expectations. Stress testing should therefore use narrative scenarios, not only statistical shocks. One scenario might assume a 35% decline in AI-infrastructure equities, a doubling of margin requirements, a 50% reduction in trading volume and a six-month delay in private financing. The important question is not whether the scenario is likely. It is whether the fund survives it.

Liquidity-adjusted sizing is equally important. A manager who believes a stock is worth twice its market price may want a large position. The position should still be capped by the amount that can be sold without threatening the portfolio. Conviction determines whether to own the asset; liquidity determines how much can be owned safely.

Leverage limits also need to respond to gains. If a fund’s equity triples, a fixed borrowing multiple permits a much larger dollar portfolio. That may seem prudent because the ratio is unchanged, but the market may not have enough depth to support the new scale. Capacity is therefore not merely a leverage ratio. It is a function of average daily volume, free float, counterparty concentration and the ability to hedge.

The Mental-Health Test Is Crude—but Financially Meaningful

Novogratz argued that a trader who cannot sleep probably has too much risk. This is not a substitute for formal risk measurement, but it captures a genuine behavioral signal. Excessive position size changes decision-making. Managers monitor prices constantly, interpret new information defensively and become reluctant to realize losses because doing so would damage identity as well as performance.

Success can make the problem worse. A manager who has produced four-digit cumulative returns may come to view volatility as temporary noise and skeptics as people who failed to understand the thesis. Investors reinforce that confidence by adding money. Employees are rewarded for supporting the winning strategy. Lenders compete to finance it. The organization can lose the internal dissent needed to challenge position size.

The remedy is governance, not personality diagnosis. Independent risk officers need authority to reduce exposure. Valuation committees need credible methods for private assets. Counterparty data should be consolidated so management can see aggregate leverage. Investors should receive sufficiently frequent reporting to understand changes in concentration and liquidity. Compensation should not reward only upside while externalizing tail risk to clients and creditors.

Small teams face particular challenges. A brilliant analyst can research companies deeply, but portfolio management, operations, compliance, treasury, legal work, collateral management and investor communication are distinct functions. Rapid asset growth can outpace the institution supporting it. A fund managing tens of billions of dollars in gross exposure requires systems and personnel that may take years to build.

What Investors Should Have Asked Before the Collapse

The failure has prompted criticism of the prominent investors who backed Situational Awareness. The more useful exercise is to identify the questions that due diligence should have answered.

How was the $45 billion measured?

Investors needed a reconciliation among net asset value, gross assets, gross exposure, derivative notional and private-company marks. Without that reconciliation, headline growth could obscure the amount of borrowing and the sensitivity of equity to a market decline.

What was the largest plausible margin call?

A financing schedule should show collateral by counterparty, contractual haircuts, discretionary margin rights, cross-default provisions and the effect of a volatility shock. A fund can have positive expected returns and still be unfinanceable under stress.

How liquid were the largest positions?

Managers often report the percentage of a portfolio that can be sold within one day, one week or one month. Those estimates should use stressed trading volume and account for ownership concentration. A public stock is not automatically liquid when the fund owns a large block.

How independent was the risk function?

Investors should know whether a chief risk officer could override the portfolio manager, whether limits were binding and how often exceptions were granted. A policy that can be waived whenever the manager is confident is not a limit.

How were private assets valued?

The methodology should distinguish recent third-party transactions, internal models and strategic funding rounds. Investors also need to know whether private gains were included in performance fees and whether those marks supported borrowing.

What happened if redemptions coincided with margin calls?

Liquidity terms, gates, side pockets and suspension rights determine who bears the cost when a fund cannot sell private assets. A portfolio can be economically solvent but unable to meet cash obligations on schedule.

Was the team scaled for the assets?

Operational due diligence should examine staffing, trade reconciliation, collateral systems, cybersecurity, disaster recovery and compliance. Investment brilliance does not compensate for weak infrastructure when the portfolio becomes large.

These questions do not guarantee safety. They change the probability that investors recognize a mismatch between the strategy and the fund structure before a crisis.

What the Collapse Says About Crowding in AI Stocks

Crowding occurs when many investors hold similar positions for similar reasons. It does not require identical portfolios. A long position in a memory producer, a data-center developer, a power company and a specialized cloud provider can all depend on the same expectation of uninterrupted AI capital spending. When that expectation weakens, the positions may fall together.

The July episode showed that crowding had extended beyond Nvidia and the largest technology companies. Investors had moved into second- and third-order beneficiaries: utilities, fuel cells, copper, networking, storage, land, cooling and former bitcoin miners. These trades often offered more upside because the companies were smaller. They also had less liquidity and greater sensitivity to financing conditions.

Forced selling can make crowded trades look fundamentally broken even when the news has not changed. A large block sold at a discount resets the market price for every holder. Risk models respond to the higher volatility. Other funds reduce positions to protect monthly performance. Dealers hedge options. The price decline becomes partly self-reinforcing.

The subsequent rebound after Citadel’s purchase suggests that at least part of the July move was technical. Once the forced seller was removed, buyers could re-enter. That supports Novogratz’s description of a possible tradable bottom in selected AI infrastructure names. It does not establish a durable fundamental bottom. A technical rebound can coexist with excessive valuations, weakening cash flow or future earnings disappointments.

Investors should therefore separate three questions. Is AI demand still growing? Are the companies converting that demand into durable cash flow? Are their securities priced attractively relative to the risks? Situational Awareness appears to have been highly confident about the first question and selective about the second, but its leverage left too little room for error on the third.

The Private-Asset Problem: Anthropic May Be Valuable, but It Is Not Cash

Reporting indicates that Situational Awareness retained an Anthropic stake valued near $5 billion. That holding could become the center of the firm’s recovery story. It could also complicate the assessment of what remains.

A private-company valuation normally comes from a financing round, secondary transaction or internal model. The headline valuation applies to the company as a whole, but different share classes can carry different rights. Strategic investors may receive commercial agreements, cloud commitments or governance rights that affect the economics. A small transaction at a high valuation does not guarantee that a large block can be sold at the same price.

Anthropic has strong strategic backing and growing commercial relevance. Amazon recorded large accounting gains on its Anthropic investment in 2026, including a $53.4 billion pre-tax gain primarily related to the stake in the second quarter. Microsoft also referred to a gain on an Anthropic investment in its fiscal fourth-quarter materials. Those gains demonstrate how rapidly private AI valuations can affect public-company earnings.

They also show why investors must distinguish operating performance from valuation gains. Amazon’s second-quarter net income of $62.6 billion included the large non-operating gain. The cash economics of AWS and retail operations were strong, but the private mark made statutory net income unusually high. A hedge fund with a large Anthropic position can similarly report strong net asset value without receiving equivalent cash.

If Situational Awareness faces investor redemptions, it may need to place the Anthropic stake in a side pocket, arrange a secondary sale or negotiate longer withdrawal terms. A rushed secondary transaction could occur below the last funding-round valuation. Conversely, a future IPO or higher-priced financing could materially improve the fund’s position. The range of outcomes is wide because the asset is both valuable and illiquid.

Regulatory Visibility: Why the Public Could See the Portfolio but Not the Risk

Situational Awareness was visible in SEC filings, yet the most important risk—leverage—remained largely hidden. That is a structural feature of U.S. disclosure rules.

Form 13F requires institutional investment managers above a threshold to report certain U.S.-listed securities quarterly. The filing is delayed by up to 45 days. It does not provide cash, borrowing, short positions, most swaps, financing terms, private assets or a complete view of options. The SEC explicitly warns that acceptance of a filing does not mean the agency has verified its accuracy.

Form ADV provides regulators and clients with information about an adviser’s business, assets and disciplinary history, but public data still do not reveal a live risk dashboard. Form PF gives regulators confidential information about large private funds, including leverage and exposures, yet the filings are not generally available to the public. That trade-off protects proprietary strategies but limits market discipline.

The episode strengthens the case for better aggregation of counterparty exposure. Archegos showed that individual banks may not know a client’s total leverage across competitors. Situational Awareness raises the same question in a different form. Regulators do not necessarily need to publish every position in real time, but they need enough data to identify rapid growth, concentrated financing and correlated exposures before a forced sale affects markets.

There is also a case for clearer investor reporting. Sophisticated clients can negotiate transparency beyond regulatory minimums. Monthly exposure reports, liquidity buckets, stress losses and counterparty concentration would have been more informative than a spectacular return number. Regulation sets a floor; fiduciary due diligence should demand more.

Does the Fund’s 80% Year-to-Date Gain Change the Verdict?

A fund that loses 67% in one month but remains up 80% for the year presents an unusual performance puzzle. The annual return sounds excellent. The drawdown sounds disastrous. Both can be true because percentage returns are path-dependent.

If $100 grows by 439%, it becomes $539. A subsequent 67% decline reduces it to about $178. The investor is still up approximately 78% from the starting point, close to the reported year-to-date result. Yet more than $360 of peak value has disappeared. An investor who entered near the beginning of the year has a large gain; an investor who added capital near the peak may have a severe loss. Aggregate performance can conceal radically different client experiences.

The drawdown also changes the quality of the track record. A strategy that compounds at high rates with controlled losses is more valuable than one that reaches the same ending value through extreme volatility. Investors evaluate Sharpe ratios, downside deviation, maximum drawdown and the consistency of returns because survival and liquidity matter. A 67% monthly loss implies that the strategy carried far more tail risk than the earlier return record revealed.

Fees further complicate the comparison. Hedge funds typically charge management and performance fees, though Situational Awareness’s exact terms are not fully public. If performance fees were crystallized after earlier gains, some investors may have paid fees on profits that were later lost. High-water marks can prevent new incentive fees until losses are recovered, but they do not return previously paid fees unless the agreement includes a clawback.

The 80% figure therefore does not erase the failure. It shows that the fund’s earlier success was real enough to leave substantial residual gains. It also shows why peak-to-trough loss, liquidity and risk-adjusted return are necessary alongside calendar-year performance.

What the Episode Means for Hedge Funds and Prime Brokers

For hedge funds, Situational Awareness is a reminder that capacity must be treated as a risk limit. A strategy that works at $500 million may not work at $20 billion. The manager may be forced into larger companies, more leverage, less liquid securities or derivative structures that change the original risk. Asset gathering can undermine the edge that attracted the assets.

For prime brokers, the event will likely prompt reviews of concentrated AI-related books. Banks may raise haircuts, reduce single-name limits or demand more frequent position data from fast-growing clients. Those changes can affect funds that had nothing to do with Situational Awareness. After a visible failure, risk departments often tighten terms across an entire strategy.

That collective tightening can itself move markets. If multiple funds are required to reduce leverage in the same names, selling pressure persists even after the original distressed portfolio is transferred. Conversely, well-capitalized firms may identify opportunities in securities depressed by technical selling. The result is a transfer of assets from highly levered specialists to larger multistrategy platforms.

The episode may also reinforce the concentration of the hedge-fund industry. Large firms can offer diversified capital, internal financing and sophisticated risk controls. They can hire specialists and absorb distressed books. Smaller funds may have sharper ideas but face higher financing costs and less tolerance from counterparties after a shock. That dynamic can make the biggest firms even more dominant.

What It Means for AI Companies and Public-Market Investors

For AI companies, the immediate lesson is that access to capital cannot be assumed. Infrastructure businesses often require spending years before revenue arrives. If public investors become less willing to finance that gap, companies may need more expensive debt, strategic partners or slower construction plans. The best-capitalized hyperscalers will be less affected than smaller providers.

For public-market investors, the collapse provides a framework for separating business risk from market-structure risk. A company’s earnings outlook may be unchanged while its stock falls because a large holder is liquidating. That can create opportunity, but only after investors examine balance-sheet strength, customer contracts, valuation and financing needs. Buying simply because a forced seller exists is not analysis.

The event also argues against treating “AI” as one trade. Chip designers, foundries, memory producers, cloud platforms, data-center landlords, power suppliers and application companies have different economics. Some earn high margins and generate cash today. Others are funding projects that may not produce revenue for years. Some have diversified businesses that can absorb mistakes. Others depend on a single customer or site.

Valuation discipline remains essential. A company can beat earnings expectations and still fall if the market had priced in a larger beat. It can grow revenue rapidly while destroying value if capital intensity rises faster. It can benefit from AI demand while its stock underperforms because the starting multiple was too high. The underlying technology does not repeal the relationship between price and return.

What Happens Next for Situational Awareness?

The fund’s next phase depends on four variables: remaining assets, investor behavior, counterparty support and the credibility of a revised risk process.

Remaining assets: The private portfolio, especially Anthropic, may preserve significant value. The firm will need reliable valuations and a plan for managing liquidity around those holdings.

Investor behavior: Some early investors may accept the drawdown because they remain profitable and believe in the long-term thesis. Later investors may seek redemptions. The fund’s governing documents will determine how quickly capital can leave and whether private holdings are separated into side pockets.

Counterparty support: Prime brokers that survived the event without losses may continue relationships under tighter terms. Less leverage, higher collateral and smaller position limits would reduce expected returns but improve survival.

Risk credibility: Investors will want evidence that the organization changed. That could include a stronger independent risk team, lower gross exposure, enhanced liquidity limits, more frequent reporting and clearer authority to cut positions. A promise to be more careful will not be enough.

The next regulatory filing will not provide an immediate answer. The Form 13F for June 30 is due in mid-August and captures positions before the late-July sale. It may show how large the book became at quarter-end, but not the final liquidation. The September 30 filing, due in November, should provide a better public view of the surviving reportable positions. Even then, private assets and borrowing will remain outside the picture.

Aschenbrenner is young enough to rebuild a career. Markets have repeatedly allowed talented managers to return after large failures, especially when the original thesis retains merit and clients did not suffer fraud or misconduct. The decisive question is whether the next version of the strategy treats survival as a constraint rather than an afterthought.

“1999 Capex Plus 2008 Credit”: A Useful Analogy With Important Limits

One of the sharper comparisons in the discussion was that the current AI market combines the capital-expenditure intensity of the late-1990s telecommunications boom with the credit dependence associated with the pre-2008 financial system. The phrase is not a literal claim that today’s market is repeating either period. It highlights a specific risk: long-lived infrastructure is being financed at enormous scale before the ultimate pattern of demand, pricing and competition is known.

The late-1990s fiber boom is a relevant example because the underlying forecast was directionally correct. Internet traffic did grow explosively. Global communications did require vast fiber networks. Yet companies such as Global Crossing spent heavily, used debt and encountered a mismatch between the arrival of capacity and the timing of profitable demand. The infrastructure survived and later supported valuable services, but some original equity and debt investors suffered devastating losses.

AI can produce the same distinction between technological success and financial success. A data center completed in 2027 may be heavily used for years. That does not guarantee that its owner earns the return assumed when the land, power, chips and debt were contracted. Competition may lower compute prices. New chips may reduce the amount of hardware needed per task. Customers may shift workloads to their own infrastructure. Power costs may rise. Construction delays may cause interest to accrue before revenue begins.

The 2008 part of the analogy concerns layering. The largest hyperscalers can fund much of their spending from operating cash flow, but the broader ecosystem includes equipment leases, project finance, private credit, vendor financing, power-purchase agreements and special-purpose vehicles. Each contract may be reasonable on its own. The system becomes fragile when several parties depend on the same assumptions about utilization, refinancing and residual asset value.

Situational Awareness sat above that industrial structure as a leveraged owner of securities. Its own lenders financed a portfolio whose companies often depended on external financing for their projects. This created leverage on leverage: borrowed fund capital invested in businesses undertaking capital-intensive expansion. A decline in equity valuations could therefore reflect not only lower expected profits but a higher cost of capital for the companies themselves. That feedback loop can turn a market correction into a fundamental financing problem.

The analogy has limits. Today’s hyperscalers are among the most profitable companies in history, with large cash balances, diversified revenue and direct customer relationships. Much of the late-1990s telecom buildout was undertaken by weaker companies dependent on continuous access to debt and equity markets. Bank regulation and counterparty reporting have also improved since 2008, even if private-fund leverage remains difficult for the public to observe.

AI demand is also generating measurable revenue now. AWS, Azure and Google Cloud are growing, and companies report shortages rather than empty facilities in key parts of the market. The bearish case is therefore not that all AI capacity will be stranded. It is that the marginal project and the marginal investor may earn less than expected once the highest-return opportunities are filled.

This is where disciplined capital allocation matters. A company that stages construction, secures customers before committing capital and maintains a strong balance sheet can survive a slower demand path. A company that builds speculatively with expensive debt has less room. A fund that owns both but finances the portfolio with high leverage may discover that the market treats them similarly during a selloff.

Why a Correct Thesis Needs a Financing Thesis

Investment analysis often separates company research from portfolio financing. The Situational Awareness episode shows that they are inseparable. A manager may estimate the future earnings of a data-center company with great precision and still fail if the fund’s own borrowing terms are not aligned with the investment horizon.

A complete thesis should answer three clocks. The first is the technology clock: when will AI capabilities and demand develop? The second is the corporate clock: when will the company convert spending into revenue and free cash flow? The third is the financing clock: how long will lenders and investors allow the position to remain under pressure?

Situational Awareness appears to have had a strong view on the first clock and detailed views on selected companies in the second. The third clock proved shortest. Margin agreements can demand action daily. Public-company projects may take quarters or years. Frontier AI development may unfold over a decade. A portfolio survives only if its capital structure bridges those timelines.

Permanent capital is one solution. A closed-end fund, insurance balance sheet or holding company can tolerate volatility without redemptions, provided it avoids excessive debt. Lower leverage is another. Longer-dated financing can help, although lenders charge more for certainty. Diversification across independent return drivers can reduce the probability that all collateral weakens at once.

None of these choices is free. Permanent capital may be difficult to raise. Lower leverage reduces upside. Long-term financing is expensive. Diversification can dilute the best ideas. Risk management is therefore not an attempt to eliminate volatility; it is the process of paying enough return to preserve the ability to continue.

The lesson applies beyond hedge funds. A household can be right that a home will appreciate over 20 years and still lose it if an adjustable mortgage becomes unaffordable. A company can own valuable assets and enter bankruptcy because debt matures before the assets generate cash. A bank can hold money-good loans and fail if depositors withdraw faster than the loans can be sold. Across finance, timing and liquidity determine whether long-term value can be realized.

Could the July Sale Mark a Bottom for AI Infrastructure Stocks?

A forced liquidation often creates the conditions for a short-term bottom. Prices fall for reasons unrelated to new fundamental information, the seller becomes exhausted and a stronger holder acquires the assets. Once the market knows the overhang is gone, traders cover shorts and long-term buyers return.

Several features of the Citadel transaction support that interpretation. The portfolio was transferred in a package rather than dripped into the market. Citadel had the capital to hold or hedge positions. Reporting linked the transaction to rebounds in selected stocks and to strong July performance at Citadel funds. The removal of an urgent seller can materially improve market microstructure.

A durable bottom requires more. Companies must meet project milestones, customers must honor contracts and capital spending must translate into revenue. Valuations must offer adequate returns after accounting for dilution, debt and future capital needs. If those fundamentals deteriorate, the absence of one forced seller will not prevent further declines.

The distinction between a tradable bottom and an investment bottom is therefore useful. A tradable bottom is defined by positioning, liquidity and near-term price behavior. An investment bottom is defined by a favorable relationship between long-term cash flow and price. Situational Awareness’s sale may have created the first in some securities. It does not prove the second.

Public investors should also be alert to secondary effects. Prime brokers may tighten financing for funds holding the same names. Companies whose shares fell may face more expensive equity issuance. Employee compensation tied to stock prices may become less attractive. Suppliers and customers may renegotiate. Technical selling can eventually influence fundamentals through the cost of capital.

At the same time, the strongest companies can use a downturn to consolidate. Well-funded cloud platforms can negotiate better equipment and power contracts. Data-center operators with committed customers can acquire distressed sites. Larger hedge funds can provide liquidity to smaller managers. The July episode may therefore accelerate the transfer of AI infrastructure from speculative capital to better-capitalized owners.

Frequently Asked Questions

What happened to Situational Awareness?

The hedge fund lost approximately 67% in July 2026 after AI-related stocks fell and lenders issued margin calls. It sold most of its public-equity portfolio, valued at roughly $16 billion, to Citadel. The firm continued operating with remaining private investments and other assets.

Did Situational Awareness lose $45 billion?

That has not been established. The $45 billion figure appears to describe peak assets, portfolio value or gross exposure that included leverage. It is not a verified measure of investor cash lost. The most reliable public performance figure is the reported 67% decline in July.

Was the fund bankrupt?

No public bankruptcy filing has been reported. The fund experienced a liquidity and margin crisis, completed a large asset sale and retained private holdings. Financial distress and forced deleveraging are not the same as legal bankruptcy.

Who is Leopold Aschenbrenner?

Aschenbrenner is a former OpenAI Superalignment researcher and the founder of Situational Awareness. He gained prominence after publishing a 2024 essay series predicting rapid advances in AI and a large industrial buildout around compute, power and data centers.

Why did Citadel buy the portfolio?

Citadel could absorb the risk and reportedly acquired the positions at a discount exceeding 10%. A large, diversified buyer can hedge, distribute and unwind positions more patiently than a fund facing immediate margin calls.

Was Situational Awareness simply long AI stocks?

No. The March 2026 Form 13F showed long positions in AI infrastructure companies and large put-option entries linked to semiconductors and technology. The exact net risk cannot be reconstructed because the filing omits strike prices, option premiums, borrowing, shorts and many non-reportable positions.

How could a 67% loss leave the fund up 80% for the year?

The fund had reportedly gained more than 400% earlier in 2026. A large percentage loss from a much higher peak can still leave the ending value above the year’s starting point. Investors who entered at different times may have very different returns.

Is this proof that the AI bubble has burst?

No. The event proves that a leveraged, concentrated AI-related portfolio was vulnerable to a sharp reversal. Major technology companies continue to report strong cloud growth and extraordinary infrastructure spending. Whether individual AI securities are overvalued depends on their cash flow, capital needs and price.

Was this another Long-Term Capital Management crisis?

There are similarities in leverage, correlated losses and forced liquidation, but the systemic impact was much smaller. LTCM had enormous global derivatives exposure and required a private-sector recapitalization organized through the New York Fed. Situational Awareness’s public book was transferred to Citadel without a comparable regulatory intervention.

What is the biggest lesson for investors?

A long-term thesis can be correct while the position still fails. Leverage, liquidity and financing determine whether an investor can remain in the trade long enough for the thesis to work.

When will new SEC filings show the sale?

The mid-August 13F covers June 30 and will predate the sale. The filing for September 30, normally due in mid-November, should better reflect the reduced public portfolio, though it will still exclude private assets, borrowing and many derivatives.

Could the fund recover through Anthropic?

Potentially. A large Anthropic stake could appreciate if the company continues growing or completes a successful financing or IPO. The position is illiquid, its public valuation is approximate and it cannot be treated as cash available on demand.

Final Assessment

Situational Awareness did not fail because it noticed the wrong economic trend. The fund identified genuine constraints in the AI buildout and invested in companies exposed to power, memory, data centers and specialized compute. Those ideas produced extraordinary gains and attracted sophisticated backers.

It failed in July because the portfolio structure could not tolerate a sharp reversal. Leverage multiplied the losses. Concentration made the positions move together. Limited liquidity weakened the fund’s bargaining power. Margin calls transferred control of the holding period to lenders. The resulting sale gave Citadel the discount and the time horizon that Situational Awareness no longer possessed.

The strongest supportive interpretation is that the core AI infrastructure thesis remains intact. Big Tech’s spending is measurable, cloud demand is growing and private AI companies may create substantial value. Selected public holdings may have been pushed below fundamental value by forced selling.

The strongest concern is that the buildout is becoming more capital-intensive and financially interconnected while token prices, hardware efficiency and competitive dynamics remain uncertain. Companies and funds are making long-duration commitments against revenue streams that may change rapidly. When those commitments are financed with leverage, even a temporary reassessment can become a solvency or liquidity event.

The event should therefore be read as a warning about financial architecture, not a verdict on artificial intelligence. The question for investors is no longer only who will benefit from AI. It is who can finance the buildout, earn an adequate return on the capital and survive the inevitable periods when market prices move faster than the technology.

Sources

• Reuters: Situational Awareness portfolio sinks 67% in July
• Reuters: Citadel funds gain after AI selloff
• Financial Times: Citadel buys Situational Awareness equity holdings
• The Wall Street Journal: Situational Awareness down 67% in July
• The Wall Street Journal: Investors that backed Situational Awareness
• SEC: Situational Awareness Form 13F for March 31, 2026
• SEC: Situational Awareness Form 13F information table
• SEC: Frequently asked questions about Form 13F
• Leopold Aschenbrenner: Situational Awareness—The Decade Ahead
• OpenAI: Introducing Superalignment
• Office of Financial Research: Hedge-fund leverage and underlying risks
• Federal Reserve: May 2026 Financial Stability Report, accessible version
• Federal Reserve: July 2026 Monetary Policy Report
• Federal Reserve testimony on Long-Term Capital Management
• U.S. Government Accountability Office: Long-Term Capital Management report
• Credit Suisse filing on Archegos-related losses
• Alphabet Q2 2026 earnings call
• Microsoft fiscal Q4 2026 earnings materials
• Amazon Q2 2026 results
• Meta Q2 2026 earnings release

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.