SEC Subpoenas Ask How Banks Lent to a 400%-Leveraged AI Fund—Is a Mandated-Deleveraging Rule Next?

In July, a 24-year-old former OpenAI researcher ran a hedge fund that at its peak had roughly $45 billion of gross exposure, borrowed at about four times its investors' money, and concentrated it in a handful of AI infrastructure stocks. When those names turned, the fund lost about 67% of its value in a month and, with hours to spare, sold its entire public portfolio in a single overnight block to Ken Griffin's Citadel at roughly a 10% discount. Now the SEC has subpoenaed Goldman Sachs, JPMorgan, Citigroup, and Bank of America — asking how they lent to this fund, how much leverage they allowed, and, pointedly, when they made their margin calls.

The question everyone is asking is whether this is a one-off investigation or the discovery stage of something bigger: formal SEC rule-making that would cap how much leverage prime brokers can hand to highly levered funds. That distinction matters for a specific reason. A mandated margin or leverage rule would permanently raise the cost of borrowed capital for these funds, forcing them to unwind crowded AI and semiconductor positions, while investors who hold the same stocks without leverage would sit mostly untouched.

Before deciding which way this goes, it helps to understand which direction the loss actually flowed. It did not flow to the banks.

A prime broker is a Wall Street bank that lends a hedge fund money and securities to bet with. The collateral for that loan is the fund's own positions, and because the loan is collateralized the bank gets to set the terms — most importantly the margin, the cushion between what the positions are worth and what the fund has borrowed. When positions fall and the cushion shrinks, the bank issues a margin call: post more collateral or we sell. That is a private contract between the fund and its lender, not a statute; the bank's margin is a risk decision, not a regulated number.

Here, that private machinery worked exactly as designed, and the banks came out clean. The fund "never missed a margin call," by one account, and the prime brokers "never risked a credit loss" — dynamic margining automatically delevered the book as AI shares ran up, and the banks had recourse to the fund's private holdings, including a stake in Anthropic. Bank of America's CEO, Brian Moynihan, said his bank "would have been fine" even without the Citadel rescue. The 67% loss did not land on the lenders. It landed on the fund's own investors, the equity that sits behind the debt. Margin calls did not move the risk to the banks; they moved the losses onto the fund's capital — which is exactly what a margin call is for.

That is the strangest and most important fact in the story, because it is the thing a mandatory-margin rule would usually be born from, and it did not happen here.

The history rhymes. In 2021, Archegos Capital — another concentrated, borrowed-up fund, this one using total-return swaps — blew up and cost two banks, Credit Suisse and Nomura, billions. That was a true failure of the mechanism: the lenders themselves took the hit. The regulatory aftermath, though, was not a new margin rule. There was no statutory cap on prime-broker leverage, no mandatory margin percentage imposed on funds. What changed was internal: banks tightened their own risk management and, in the end, enforcement — Archegos founder Bill Hwang was convicted of fraud in 2024. Prime-broker margin stayed a private credit decision.

The Situational Awareness probe, so far, looks like the same shape. The SEC has described its request as information-gathering, with no wrongdoing alleged — the posture of an enforcement inquiry, not the opening of a rule. And this SEC, run by Chair Paul Atkins, has published a 2026 regulatory agenda that reads as a deregulatory roadmap and contains no item for hedge-fund leverage or prime-broker margin. Add the fact that the lenders here lost nothing, and the case for a new mandate looks thin: rules tend to follow when a mechanism fails for the institutions it is meant to regulate, and this mechanism held.

But "no new statute" is not the same as "nothing structural is happening." The thing the subpoena is really exposing does have a discovery stage, and it is up in the plumbing, not in a rulebook. Prime-broker financing is highly concentrated — one analysis puts about two-thirds of the market in four banks — and hedge-fund gross leverage has climbed to levels not seen since before the global financial crisis. When a handful of lenders share the same big clients, their margin calls become correlated: they read the same models and the same signals, so one bank tightening reinforces the next. That procyclical dynamic is the actual structural risk, and no subpoena or margin cap fixes it directly.

Here is where the leverage economics bite. This dynamic — whether it arrives as a formal rule or, more likely, as looser self-regulation — raises the cost of leverage for the funds that depend on it first. The funds hurt soonest are concentrated, borrowed-up AI and semiconductor funds that need the financing to express their thesis; funds relying on lines the banks can terminate on short notice rather than committed term funding; funds whose positions have already run up sharply, because banks raise haircuts after big run-ups (Moynihan described tightening "the underwriting standards, just a hair" precisely for that reason); and funds tied to a single prime broker that have no second lender to fall back on. An unleveraged shareholder in the same AI infrastructure names needs none of that. Higher margin requirements reduce how much leverage each dollar of fund capital buys, which drags down returns for a strategy whose whole point was to multiply them.

So which way is this really going? The falsification test for the rule-making thesis is largely already met: a deregulatory agenda with no such rule, an investigation still in its information-gathering stage, a precedent where a far costlier failure produced tightening instead of a statute, and no lender losses here to create the political demand for one. The market is, in effect, already writing the margin rule itself — the dynamic margining that threw the fund's own investors under the bus, the haircuts on the winners, the pulled lines. What would genuinely revive the rule thesis is a specific finding: evidence that the same banks that lent also timed their margin calls, or coordinated the block sale to Citadel, in a way the SEC decides was abusive rather than just fast. That, and not the subpoena itself, is the signal worth watching.

The margins that gutted this fund were not written by any regulator. They were clauses in a contract between a young man with an AI thesis and the banks that lent him four times his money. A rule would move that decision from the contract into the statute. On the evidence so far, the more likely change is what has already begun: the banks tightening their own plumbing a hair, fund by fund, and the cost of being levered into crowded AI names going up quietly from the inside.