
Situational Awareness: Goldman's $200m in Fees and the Archegos Comparison
The Financial Times says Goldman Sachs earned more than $200 million in 2026 fees from Situational Awareness, more than from Citadel or Millennium. Credit Suisse made $17.5 million from Archegos and lost $5.5 billion. The desk expects Goldman to report no Situational Awareness loss and more than $3 billion of equities financing revenue on October 13.
In the spring of 2021 Credit Suisse's prime brokerage desk learned what a hedge fund client is worth. The bank had earned $17.5 million from Archegos Capital Management in the year before the family office defaulted, then lost about $5.5 billion on it. That is roughly 314 dollars lost for every dollar of revenue.
Goldman Sachs has the opposite ledger. The Financial Times reported on Thursday that the bank earned more than $200 million in fees this year from Situational Awareness, the AI hedge fund run by 24-year-old former OpenAI researcher Leopold Aschenbrenner, making it the largest client in Goldman's prime brokerage business, ahead of Citadel and Millennium. Two months ago that client was forced to sell its public equity book to Citadel after margin calls. Goldman kept the fees and, on everything reported so far, lost nothing. The desk's view is that this is the correct reading of the episode: the prime brokers priced the risk well and got out first, and the losses landed on the fund's investors. The regulatory question is what happens when the next block has no Citadel to take it.
The rise and fall took about two years. Situational Awareness launched in 2024 on the thesis of Aschenbrenner's essay of the same name: AI would need vast amounts of compute, memory and power. Its Form ADV showed about $9.3 billion of regulatory assets at the end of 2025. It returned 439% net in the first half of 2026. CNBC reported that the fund commanded roughly $45 billion at its July peak and about $10 billion after the sale, with CNBC's David Faber putting gross leverage at as much as four times. Reuters, counting differently, put prior assets at about $20 billion. The gap between the two numbers is largely borrowed money.
Four times leverage explains the speed. At 4x, a 17% fall in the long book wipes out 68% of equity before hedges. Situational Awareness lost about 67% in July, the Wall Street Journal reported. Its core holdings fell between 27% and 54% that month as the Nasdaq 100 dropped more than 10% and Korea's Kospi lost about a third. Its second-quarter 13F, filed in August, showed $20.2 billion of reportable US holdings at June 30, with SanDisk at $5.7 billion and Micron at $5.6 billion: about 56% of the reported book in two memory-chip stocks. The filing omitted the SK Hynix position listed in Seoul and the short book, so the real concentration in the memory cycle was higher. Micron fell 39% from its late-June peak.
Bank of America, Goldman and JPMorgan called for collateral. On July 30 the brokers, with Citigroup, shopped the book before the open, and Citadel bought most of the public portfolio, about $16 billion by the FT's count, with Millennium bidding against it. The discount was reported at about 10%. The fund kept its private stakes, chiefly Anthropic. The stocks Citadel bought rallied 20% to 27% by mid-afternoon that day.
What $200 million buys
Banks do not disclose client-level revenue, so the FT figure is the only look inside. It can be tested against Goldman's own numbers. Goldman reported record equities financing revenue of $3.26 billion in the second quarter, up 91% from a year earlier, and about $2.61 billion in the first. Against that $5.87 billion half-year, $200 million from one client is about 3.4%. That is large for a single name and small for the franchise.
The fee also says something about the size of the loans. The book was sold at the end of July, so the fees accrued over roughly seven months, which annualises to about $343 million. If Goldman earned a full percentage point over its own funding cost, a rich spread for a client this size, it would have needed average balances of about $34 billion to generate that. That is three quarters of the fund's reported peak, which was shared among at least three prime brokers, and far above where the book stood in January. The arithmetic only closes if Goldman carried much more than a third of the exposure, or charged well above a percentage point through swap spreads, option premiums and commissions, or both. Either way, Goldman was paid handsomely for the risk, which is what Credit Suisse failed to do in 2021.
The mechanics also differed from Archegos. The SEC's complaint said Archegos grew from $1.5 billion with $10 billion of exposure to more than $36 billion with $160 billion of exposure, about 4.4 times, largely through total return swaps that each bank saw only in part. Archegos was a family office that filed nothing. Goldman sold early and came through largely unscathed; Morgan Stanley lost about $911 million and Nomura about $2.9 billion. Situational Awareness is a registered adviser that files 13Fs and Form PF, its leverage was known to the market, and its brokers coordinated a single block sale in hours. The 2021 lesson, that the first bank to the exit wins, had been learnt by every bank at the table.
The case against, and the regulators
The system worked because Citadel had the appetite and the balance sheet on one Thursday morning. Situational Awareness was not alone in these trades. Asia-focused fundamental long-short funds were down 18.6% in July through the 28th, according to a Goldman prime brokerage note cited by Reuters. If two or three leveraged funds with the same memory and data-centre names had needed buyers that week, the discount would have been far wider than 10% and the collateral could have fallen short of the loans. Crowding across many clients in the same names is how a hedge fund loss becomes a bank loss.
That risk is real, and supervisors have started looking. The FT reported this week that the Federal Reserve and the Bank of England's Prudential Regulation Authority are examining banks' exposures to large trading firms, including Jane Street, one of Situational Awareness's investors, and how those exposures moved during the trading day. The SEC is moving in the other direction. In April it and the CFTC proposed amendments to Form PF that would drop the requirement for large hedge fund advisers to file a current report within 72 hours of a margin default, and would raise the large-adviser threshold from $1.5 billion to $10 billion of hedge fund assets. A fund the size Situational Awareness is today would sit on that line. On September 3 the agencies pushed compliance with the 2024 Form PF amendments to July 1, 2027. Form PF data are confidential in any case; they tell the SEC what a fund owes, not what the next buyer can absorb.
The winners are easy to name. Citadel bought a book at a discount and watched it rally. Goldman, JPMorgan and Bank of America were paid through the rise and were fully collateralised on the fall. Clear Street, which the FT reported this month is now the fund's prime broker, has a high-profile client. The losers are the limited partners who were up 439% at the end of June and down 67% a month later, and the funds that bought the memory trade late. Aschenbrenner himself is buying again: CNBC reported in September that the fund had paid hundreds of millions of dollars in option premiums on names including SK Hynix, SanDisk, AMD, Bloom Energy and CoreWeave. Options cap the loss at the premium, so the next margin call, if it comes, cannot reach a broker.
The desk expects Goldman to report third-quarter equities financing revenue above $3 billion on October 13, and to disclose no credit loss tied to Situational Awareness. A loss provision or a charge in equities financing would break the call and point to more exposure than the FT's fee figure suggests. The next check is the fund's third-quarter 13F, due November 16, which will show how much of the rebuilt book sits in the same memory names. The SEC's final Form PF rule, due before the July 2027 compliance date, will show whether the margin-default report survives.