Hedge fund's 439% return should have been the warning

Former OpenAI employee Leopold Aschenbrenner sent a letter to investors in his artificial intelligence-focused hedge fund, Situational Awareness, on July 24. He reported a 439% net return for the first half of the year, adding a postscript: This seemed like a good time to add funds. Six days later, Ken Griffin’s Citadel bought a chunk of the fund’s stock holdings — a stake once estimated at $16 billion — in one of the largest rushed equity transactions in Wall Street history.

The easy story is another tech wunderkind flying too close to the sun. The more useful one is that the 439% return was a warning, not a triumph.

Mathematics sets a speed limit on how fast a portfolio can compound. If you have an edge, Bell Labs scientist John Kelly worked out in 1956 the optimal amount of leverage to deploy. Bet more or less and you leave money on the table. Bet twice as much or more and you guarantee losing everything — the only question is when.

Suppose a concentrated AI portfolio really did offer a 40% annual edge over cash, on 50% annual volatility, a level of optimism that requires extreme situational awareness. Optimal leverage is then about 1.6 times, and the fastest such a fund can be expected to compound is 17% in a half year. Such a fund will have wild ups and downs, of course, doubling in six months about once every three years, tripling about once in 10 years. But a 439% return happens less than once in a century.

You cannot get there in your fund’s second year by being smart; you need luck. And lucky fools who bet too much are far more common than lucky situationally aware traders. No leaked risk reports are needed to establish that the firm, which the Financial Times reported as having seven investment professionals and 20 employees, had nobody whose job it was to say how much leverage was too much.

Nor was the fund an outlier. In South Korea, margin loans hit a record 38 trillion won ($26.4 billion) in June as retail investors borrowed to concentrate in Samsung Electronics Co. Ltd., SK Hynix Inc. and the memory complex. The Kospi benchmark equity index more than doubled in the year to June 22; it has lost roughly a third of its value since. Many of those retail accounts met their margin calls the same week Aschenbrenner met his.

A household in Seoul levered into two chip stocks and a $20 billion fund levered into a dozen AI names are the same position in different wrappers: A thesis that sounds right, exposure to a single theme, borrowed money, and no one standing between conviction and size.

That last element is the real novelty of this cycle. The risk intermediation that is supposed to be built into the system — the broker who limits margin, the prime broker who questions concentration — stood aside on the way up and protected themselves on the way down.

A few years back, investor Richard Dewey and I published "Toil and Trouble, Don’t Get Burned Shorting Bubbles," in the Journal of Investment Management, about the last time everyone knew what was coming, the short subprime-mortgage-backed-securities trade of 2005 to 2008.

Our finding was that situational awareness was the easy part. The celebrated shorts paid years of negative carry, depended on dealer marks that nearly broke some of them, and bore counterparty risk to the very banks whose failure they were betting on. Michael Burry, made famous in The Big Short, came within an investor revolt of being forced out of the greatest trade ever.

The durable fortunes were made after the crash, by buyers of the wreckage. David Tepper’s hedge fund returned 120% in 2009 buying bank stocks. Dan Ivascyn’s team bought the subprime bonds everyone else was dumping and grew Pacific Investment Management Co.’s Income Fund from $223 million to $121 billion.

The lesson was that structuring a trade — the sizing, the carry, the time horizon, the leverage — is what makes fortunes, not being right.

Citadel completes the lesson. The fund named for knowing the future was sold, at forced-sale prices, to a firm built around not needing to know it. Citadel carries plenty of leverage, almost certainly more gross exposure than Situational Awareness ever ran. The difference is that it is spread across thousands of positions under central risk limits, with portfolio managers cut quickly when losses breach small thresholds.

Citadel took this course the expensive way, after its two largest funds lost roughly half their value in 2008, surviving largely because investor money was locked up. It spent the next 18 years institutionalizing the lesson, and this week it was the one writing the check at the bottom. Showing up as the buyer when someone else must sell is not luck; it is the compounded return on two decades of boring discipline.

Situational Awareness’ unwind itself worked: billions of dollars of concentrated stock moved to one buyer inside 24 hours with no visible contagion. This suggests that the banks too learned their lesson from the losses some took in the blowup of Bill Hwang’s Archegos Capital Management in 2021. This time the machinery functioned on the way out. Why it functioned so smoothly on the way in is a question the regulators might pursue with Aschenbrenner’s prime brokers.

There is a coda. Situational Awareness will continue as a private investment firm, holding among other things a $5 billion stake in Anthropic PBC. The main position that survived is the one leverage could not reach: illiquid, unmarginable, impossible to be forced out of. No prime broker will lend against private shares with transfer restrictions, no liquid markets and no daily marks, so no prime broker could take it away.

Aschenbrenner may yet be proved right about AI’s potential to reshape the future. If the boom resumes, the gains on his old portfolio will accrue to Citadel’s investors at his investors’ expense. That is the tuition. Markets do not pay for knowing the future. They pay for being solvent when it arrives.

This column reflects the personal views of the author and does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners. Aaron Brown is a former head of financial market research at AQR Capital Management. An active crypto investor, he has venture capital investments and advisory ties with crypto firms. He is the author of "Wrong Number."