NextFin News - The U.S. Securities and Exchange Commission has sent subpoenas to four of Wall Street's biggest banks seeking information about the trading of Situational Awareness, the AI-focused hedge fund that nearly collapsed last month after a $35 billion wipeout. The subpoenas went to Goldman Sachs, JPMorgan Chase, Citigroup, and Bank of America, and they ask for the timing of the fund's trades and its communications with lenders about the borrowed money that supercharged its bets. The fund itself has not been accused of wrongdoing, and the investigation is at its earliest stages.
The subpoenas mark the first formal regulatory step into an episode that has become a stress test for the leveraged AI trade. They shift the spotlight from the 26-year-old manager who built the fund to the banks that fed it credit - and to the question of whether prime brokers understood how concentrated, how crowded, and how fragile the biggest bet in the market had become.
The Subpoenas: What Regulators Are Asking For
The SEC's outreach, first reported on August 24 and corroborated by multiple outlets, seeks two things above all: when Situational Awareness executed its trades, and what it told lenders about the leverage it was using. Regulators also warned the banks to preserve all records connected to the San Francisco-based fund. Spokespeople for all four banks declined to comment, as did the SEC.
The legal instrument matters. A subpoena is compulsory process, not a voluntary request. It signals that the agency wants documents and testimony it could later use in an enforcement action - even though no such action is assured, and even though the fund has not been charged with misconduct. The specific theory of the case - market manipulation, disclosure failures, prime-broker conduct, or something else - has not been made public.
Situational Awareness struck a cooperative tone. In a statement to reporters, a fund spokesman said:
It is to be expected that regulators would closely examine any funds that are high profile, produce significant returns or have particularly dramatic drawdowns. We are a highly regulated business and will cooperate to the fullest extent with any regulatory request.
That framing - drawdowns invite scrutiny - is accurate as far as it goes. The SEC has a well-established practice of opening files after large, sudden losses, and most such inquiries never lead to enforcement. But the scope of the document requests suggests this inquiry is about more than the fund's investment judgment.
How a $45 Billion Fund Fell to $10 Billion in Days
The backdrop is a collapse that moved with unusual speed. Founded in 2024 by Leopold Aschenbrenner, a former OpenAI researcher who turned a widely read essay on artificial general intelligence into seed capital, the fund raised roughly $225 million from Stripe co-founders Patrick and John Collison, former GitHub CEO Nat Friedman, and investor Daniel Gross. From launch through June 2026, returns exceeded 1,000%.
By early July, the fund commanded about $45 billion in assets. It did not get there on equity alone: it used as much as 400% leverage, meaning borrowed money made up the bulk of its buying power. Its public book was a concentrated bet on AI infrastructure. At the end of June, SanDisk and Micron Technology together accounted for more than 56% of its disclosed U.S. holdings. When the AI trade turned in late July, the concentration became a trap. SanDisk fell nearly 47% in July; Micron dropped roughly 29%. Bloom Energy, Taiwan Semiconductor Manufacturing, and Nebius Group - other concentrated positions - also declined.
Margin calls followed. Prime brokers Bank of America, Goldman Sachs, and JPMorgan Chase worked with the fund to meet collateral demands, but the positions could not be held. Citadel, run by Ken Griffin, purchased the leveraged public stock book at a discount, including hard-hit names such as SK Hynix and CoreWeave. By the end of the week, the fund's holdings had fallen to roughly $10 billion - a loss of about $35 billion in assets.
Griffin later told investors that Citadel had unwound the bulk of the risk it took on, completing more than 100 block trades worth upward of $4 billion in market value and cutting its exposure by more than 80%. What the fund kept was its private stake in Anthropic, the AI lab now eyeing an IPO at a valuation that could reach $2 trillion. The connection is personal as well as financial: Aschenbrenner's wife serves as chief of staff to Anthropic CEO Dario Amodei.
The Market Reaction: A Selloff That Exposed Thin Liquidity
The July selloff in the fund's holdings was not an isolated event; it was part of a broader pullback in AI-linked memory and chip names after an extraordinary first half of 2026. Since the calendar flipped to July, Micron fell roughly 20% and SanDisk was down more than 35% before the steepest losses hit, according to market commentary at the time. The Philadelphia Semiconductor Index dropped in sympathy, dragging peers such as SK Hynix and Seagate lower.
The significance is not just the direction of travel but the speed. A concentrated, leveraged book does not unwind gracefully; it gaps. When a fund holds more than half its disclosed portfolio in two names and those names fall together, there is no natural buyer on the other side - everyone who wanted exposure was already in the trade. That is why the distressed sale to Citadel was necessary: the open market could not absorb the liquidation without pushing prices to levels that would have triggered still more margin calls.
For the broader market, the episode served as a reminder that liquidity in the most crowded AI names is contingent, not guaranteed. It exists while positions are small relative to daily volume; it vanishes when a large holder needs out. The $45 billion fund learned this in days. The regulators now want to know whether the lenders priced that liquidity risk into the credit they extended.
The Real Target May Be the Lenders, Not the Fund
Here is the question the subpoenas really ask: who bears responsibility when a leveraged bet this large blows up?
On the surface, the obvious culprit is the fund's own concentration. Putting more than half of a disclosed book into two memory names while running 400% leverage is a risk decision, and it went wrong. But the SEC's document requests point in a different direction - toward the plumbing. Regulators want trade timestamps and lender communications. That is the language of an inquiry into how the trade was financed, monitored, and unwound, not merely into whether the fund's investment thesis was sound.
The mechanism is straightforward and unforgiving. Leverage amplifies gains on the way up and forces sales on the way down. When collateral values fall, prime brokers issue margin calls. If the borrower cannot post more collateral, the broker sells - and in a concentrated, crowded position, forced selling pushes prices down further, triggering more margin calls. This is not a bug in the system; it is the system working as designed. The policy question is whether the banks that extended the credit understood the size of the position, the crowding in the names, and the liquidity of the exit.
There is precedent for regulators looking past the fund to the lenders. In March 2021, Archegos Capital Management defaulted on margin loans used to acquire total return swaps, prompting lenders to dump roughly $20 billion of securities. The combined losses at banks linked to Archegos reached about $10 billion, with Credit Suisse taking roughly $5.5 billion and Nomura about $2.9 billion. The aftermath established that regulators will examine whether banks had adequate risk controls over concentrated, highly leveraged clients - and whether they were paid enough to take that risk.
The difference here is structural. Archegos used derivatives to conceal the true size of its exposures from its lenders. Situational Awareness held its biggest bets in plain sight, in 13F filings that any market participant could read. The concentration was not hidden; it was disclosed. That makes the failure less about concealment and more about a market-wide decision to look at the same numbers and decide the risk was acceptable - right up until it was not.
Is This Cyclical or Structural? A Warning Shot That May Change the Plumbing
The cyclical read is simple: this was one fund, one manager, one crowded trade. Concentrated leveraged blowups happen in every cycle - Long-Term Capital Management in 1998, Archegos in 2021, and countless smaller funds in between. Markets absorb them, leverage migrates to the next hot theme, and the plumbing is left largely intact. On this view, the subpoenas are routine post-mortem hygiene, and the AI trade will re-leverage once volatility fades.
The structural read is more consequential, and the evidence for it is building. Bank of America CEO Brian Moynihan, whose bank was one of the fund's prime brokers, called the near-collapse a "warning shot" for financial markets fueled by elevated valuations and borrowed money. His comments suggested that Wall Street's largest prime brokers are already reexamining their exposure to highly leveraged investment firms. When the biggest lenders start tightening terms, the cost of leverage rises for everyone - not just the fund that blew up.
Three pieces of evidence point to a structural shift rather than a one-off. First, the size: $45 billion concentrated in AI infrastructure is not a niche position; it is a market-wide risk factor. Second, the speed: a $35 billion loss in days shows how thin liquidity has become in the names everyone owns. Third, the timing: the collapse came as AI valuations had already stretched to levels that required uninterrupted bullish news. A market built on borrowed money and unanimous positioning has less shock absorption than it believes.
The judgment: the fund's collapse is cyclical - a single concentrated bet that mean-reverted violently. But the regulatory and lender response is structural. If prime brokers raise margin requirements, demand more collateral, or reduce concentration limits for AI-focused funds, the cost of the AI trade rises permanently. That is a slower, quieter repricing than a fund blowup, but it cuts deeper.
The Counter-Thesis: This Is Routine, and the Market Has Already Moved On
The strongest case against the structural read is that the market has digested the episode without contagion. Citadel unwound more than 80% of the risk it assumed, absorbing the distressed book through more than 100 block trades without destabilizing the names it sold. The four banks under subpoena are the most capitalized in the U.S. financial system; a document request is not a loss. And the fund's surviving asset - the Anthropic stake - could still prove enormously valuable if the IPO proceeds.
On this view, the subpoenas are the SEC doing what it always does after a large, sudden loss: open a file, collect documents, and close it if nothing illegal appears. The agency has a well-established practice of scrutinizing firms involved in sharp drawdowns, and most such inquiries never lead to enforcement.
The counter-thesis is plausible but incomplete. It treats the subpoenas as backward-looking when the lender behavior they may trigger is forward-looking. The falsifying signal is specific: if, by the next quarterly earnings season, the four subpoenaed banks report stable or rising prime-brokerage leverage books and unchanged margin terms for concentrated single-theme funds, the structural-tightening thesis is wrong. If instead they disclose tighter concentration limits, higher haircuts, or reduced risk appetite for AI-levered clients, the tightening thesis is confirmed.
The Prime-Broker Business Model Under the Microscope
Beneath the specific facts lies a business-model question. Prime brokerage is a relationship business built on financing, securities lending, and risk intermediation - and its economics reward scale and speed. In a competitive market for hedge-fund clients, the lender that offers the most generous terms often wins the flow. That creates a collective-action problem: each bank has an incentive to extend credit up to the edge of what it can tolerate, on the assumption that it can exit first if the trade turns.
Situational Awareness tested that assumption. The fund's prime brokers did not lose capital on the public book - Citadel absorbed it. But the episode revealed how quickly a marquee client can become a distressed one, and how much of the unwind depends on finding a buyer with the balance sheet and the nerve to step in. The subpoenas ask whether the banks' risk systems saw the danger early enough, and whether their communications with the fund reflected an accurate understanding of the leverage being used.
If the answer is no, the regulatory response could reshape prime-brokerage economics: higher capital charges for concentrated exposures, stricter concentration limits, more demanding margin terms. Those costs would flow through to hedge funds as higher financing rates - a structural tax on the leveraged AI trade.
What Comes Next
Three timelines matter. In the short term, expect document production and witness interviews; the investigation is at its earliest stages and may not lead to enforcement action. In the medium term, watch the banks' risk disclosures and prime-brokerage commentary for signs of tighter leverage terms - that is where the real market impact will show up. In the long term, the Anthropic IPO remains the fund's lifeline; a successful listing at a multi-trillion valuation would retroactively validate the manager's AI thesis even as the public book failed.
The asymmetry is clear. The banks are exposed to reputational and regulatory risk, not capital loss. The fund is exposed to existential risk if Anthropic disappoints. And the broader market is exposed to a slow repricing of leverage if the warning shot is heard.
The central lesson of Situational Awareness is not that a young manager bet too big. It is that a $45 billion position in plain sight, financed at 400% leverage by the safest banks in the system, could evaporate in days - and that the regulators are now asking the lenders, not just the borrower, to explain how that was allowed to happen.
Explore more exclusive insights at nextfin.ai.
