Everything About AI
NextFin News - Goldman Sachs collected more than $200mn in fees from its prime brokerage work with Situational Awareness, the AI-focused hedge fund that nearly collapsed in July after leveraged bets on artificial-intelligence stocks soured, according to people familiar with the matter.
The figure lays bare how Wall Street's largest banks profited from the fastest-rising hedge fund in recent memory even as the fund's founder, Leopold Aschenbrenner, was forced into a fire sale that rattled global technology shares. The fees — generated from trade execution, financing and other prime services — arrived as Goldman's equities financing revenue hit a record $3.26bn in the second quarter, up 91% from a year earlier. The banks got paid while the fund lost. That asymmetry is the story.
The episode is also a stress test of a business model that has quietly become one of the most profitable on Wall Street. Prime brokerage — the suite of services that lets hedge funds borrow, trade and operate at scale — has been the engine behind Goldman's record equities performance. The question now is whether the model that produced $200mn in fees from a single client can survive the next time that client is wrong.
The Fee Machine
Situational Awareness was a major client of Goldman Sachs, JPMorgan Chase, Bank of America and Citigroup, all of which provided the prime brokerage services that allowed the fund to lever its concentrated AI bets by as much as four times, according to people briefed on the arrangements. Prime brokers charge for financing positions, executing trades, lending securities and providing the operational infrastructure hedge funds need to function. The more a client trades and borrows, the more the bank earns — regardless of whether the client's bets win or lose.
The fund's rise was meteoric. Founded in July 2024 with seed capital from Stripe co-founders Patrick and John Collison, former GitHub chief executive Nat Friedman and investor Daniel Gross, it grew from a few hundred million dollars to a reported peak of about $45bn in assets, posting returns of more than 1,000% since inception. Aschenbrenner, a former OpenAI researcher who left the company in 2024 amid a dispute over his sharing of internal information, became one of the most closely watched investors on Wall Street.
"Before long, the world will wake up," Aschenbrenner wrote in 2024. "If they are seeing the future even close to correctly, we are in for a wild ride."
That wild ride was financed by the banks. A fund that grew from roughly $200mn to $45bn in two years while running leverage of up to 4x needed constant financing, constant trade execution and constant securities lending. Every leg of that growth generated fees. When the fund was winning, the banks earned on volume. When the fund was losing, the banks earned on the frantic trading required to meet margin calls.
The Unwind
The ride ended abruptly in July. As shares of AI chip and memory makers including SK Hynix, SanDisk and Micron Technology tumbled, and short positions in software companies moved against the fund, its highly leveraged portfolio came under severe pressure. The fund fell about 67% in roughly a month, with assets shrinking to around $10bn from the peak. Several of the largest memory and semiconductor stocks were down around 50%, according to a note from Jane Street, one of the fund's investors.
Prime brokers issued margin calls, demanding additional collateral. JPMorgan ultimately cut off lending to the fund after the losses, while Goldman Sachs, Bank of America and Citigroup continued working with it to reduce positions. The fund sold the bulk of its public equity portfolio — reported to be worth about $16bn at the time — to Citadel at a discount in a transaction that temporarily stabilized the position but crystallized losses.
Ken Griffin, Citadel's founder, said the firm had "successfully shed more than 80% of the aggregate risk from the original portfolio," completing nearly 100 block trades totaling more than $4bn in market value. The trades included "the largest intraday block trades of the year in 10 different names."
The contagion spread well beyond the fund itself. Jane Street, the Wall Street trading firm that had invested in Situational Awareness, posted a roughly $15bn loss in July — its worst monthly loss ever and its first down month in about a decade. The firm still recorded more than $40bn in revenue through July, its best trading year on record, but the loss was a reminder that even the most disciplined risk managers were exposed to the AI trade's reversal. "We largely lost on the same portfolio of trades that had strong outperformance in the second quarter," the firm told employees.
Regulators have taken notice. The Securities and Exchange Commission sent subpoenas to major Wall Street banks seeking information about the fund's trading and its communications with lenders, according to people briefed on the matter. The subpoenas asked for details on the timing of the fund's trades and for its communications with lenders about the money it was borrowing, also known as leverage. The fund said it would "cooperate to the fullest extent with any regulatory request."
Why the Banks Won Anyway
The $200mn-plus fee figure underscores a structural feature of prime brokerage: the bank earns on volume and leverage, not on the direction of the client's bets. Goldman Sachs has been the standout beneficiary of the hedge fund leverage boom. Its equities financing revenue — the line item capturing prime brokerage lending and securities lending — reached $3.26bn in the second quarter, a record that exceeded the bank's total for all of 2020. Equities intermediation added another $4.16bn, lifting total equities revenue to $7.42bn, up 72% year over year.
Total net revenue at Goldman reached $20.34bn in the quarter, up 39% from a year earlier, with strength spread across trading, investment banking and financing. The equities desk pulled in more revenue than any bank has ever recorded in a single quarter, and it was the third straight quarter that Goldman's equities operation had surpassed every prior benchmark set by any bank.
The business case for prime brokerage remains as strong as ever, one senior equities banker said, even as the episode serves as a "sober reminder that this crown jewel business does come with risk." The risk is not that the bank loses money on the fund's positions — prime brokers lend against collateral and call for more when prices fall — but that a disorderly unwind leaves them holding assets they cannot sell without taking a loss.
The mechanics of that protection are precise. Prime brokers lend hedge funds money against their portfolios as collateral, typically at haircuts that widen as volatility rises. When a position falls, the broker issues a margin call; the client must post more cash or securities, or the broker liquidates. In normal markets, this transfers risk from the bank to the client. In stressed markets, it transfers selling pressure from the client to the market — and if the market cannot absorb it, back to the bank.
Here is the cyclical-versus-structural call that matters. The fee windfall is cyclical: leverage cycles revert, and the extraordinary volume generated by a crowded AI trade will normalize as positioning unwinds. Goldman's record equities financing revenue will not repeat at that pace once hedge funds deleverage and trading volumes subside. But there is a structural shift underneath it. The AI buildout has correlated hedge fund books across the industry — funds that once diversified across sectors now hold overlapping exposures to the same semiconductor, memory and data-center names. When one leveraged player is forced to sell, the same names fall in everyone else's portfolio, triggering margin calls elsewhere. The prime brokerage model was designed for idiosyncratic blowups, not correlated ones.
The Counter-Thesis
Defenders of the banks argue that prime brokerage is a low-risk, fee-generative business precisely because of its collateralized lending model. Bank of America chief executive Brian Moynihan said his bank would have been "fine" even without the Citadel rescue, which provided the capital for Situational Awareness to repay its lenders. "These are all warning shots," Moynihan said. "Valuations get out, leverage in the system gets there. You have to be careful."
That view has merit as far as it goes. The banks were repaid, the fire sale was contained, and Goldman's fee income is locked in. Bank of America would have been fine. So would Goldman, on the evidence of a quarter that produced record financing revenue and no disclosed credit losses from the episode. The collateralized model did what it was designed to do: it transferred the loss to the fund and its investors, not to the banks' balance sheets.
But it mistakes the survival of one episode for the durability of the model. The prime brokerage boom has been built on a two-year bull market in AI assets that inflated collateral values and suppressed volatility. When collateral falls and volatility spikes simultaneously — as it did in July — the margin mechanism that protects banks becomes the mechanism that amplifies the downturn, forcing simultaneous liquidations that push prices lower and trigger more margin calls. The protection works only if someone can buy.
The strongest evidence that the model is more fragile than its defenders admit is that the fund's collapse did not stay contained. Jane Street's $15bn loss, the forced sale of a portfolio reported at about $16bn, and the SEC's subpoenas to multiple banks show that the transmission channels between a single hedge fund and the broader system remain wide open. The banks may have earned their fees, but they also demonstrated that their risk controls depend on markets remaining liquid enough to sell into — an assumption that failed precisely when it mattered most.
The falsifying test is straightforward. If the banks report no material rise in prime-brokerage credit losses through the end of 2026, and if the SEC's inquiry closes without enforcement action or new margin requirements, then the fragility argument is overstated and the model proved resilient. If either condition breaks — rising credit costs or regulatory tightening — the $200mn fee windfall will look like the last easy money of the cycle.
What Comes Next
In the short term, expect prime brokers to tighten underwriting standards modestly. Moynihan acknowledged as much, saying "the tendency is to tighten the underwriting standards, just a hair." Goldman Sachs and its peers are unlikely to abandon a business that generated record revenue, but they will demand more collateral and run tighter stress tests on concentrated, leveraged AI books. The tightening will be marginal, not revolutionary — the fee economics are too attractive, and competition for prime brokerage clients remains fierce.
Over the medium term, the regulatory scrutiny is the bigger risk. The SEC's subpoenas seek details on the timing of the fund's trades and its communications with lenders. If regulators conclude that banks failed to manage their exposure appropriately or that margin practices contributed to the disorderly unwind, the industry could face new capital or reporting requirements that compress prime brokerage returns. The inquiry is at an early stage, and no enforcement action has been announced, but the mere fact of subpoenas to multiple banks signals that supervisors view the episode as systemic rather than idiosyncratic.
In the long term, the question is whether the AI trade has shifted from a momentum-driven boom to a fundamentals-driven market. Goldman Sachs research forecasts global AI-related investment approaching $1tn in 2026, and the buildout in computing power, semiconductors, memory and energy infrastructure is real. The fund's thesis was not wrong in direction. It was wrong in timing and leverage. The structural demand for AI infrastructure does not guarantee that equity prices can absorb it without the kind of leverage that turns a correction into a cascade.
The verdict: Goldman's $200mn-plus fee windfall is real money earned on a business model that worked as designed. But the design assumes orderly markets, and July showed that the AI trade's greatest vulnerability is not the thesis — it is the leverage layered on top of it. The banks got paid. The next time, they may not get out.
Explore more exclusive insights at nextfin.ai.
