The Hedge Fund That Bet Twice on the Same Future

For a few weeks in July, the clock that mattered in Leopold Aschenbrenner's life was not the one measuring the arrival of superintelligence. It was the one kept by his prime brokers. As AI-linked shares fell and software stocks rallied, lenders wanted more collateral. A thesis designed to span a decade collided with obligations measured overnight. By the time the calls stopped, Situational Awareness had sold most of its public-equity portfolio to Citadel.

The shorthand is that Aschenbrenner “lost $35 billion.” It is dramatic and arithmetically tidy: reports placed the fund near $45 billion at its early-July peak and around $10 billion after the unwind. But it is not a verified personal loss, nor necessarily a clean trading-loss number. Assets under management can change with performance, leverage, subscriptions, redemptions and private-company marks. The more useful fact is less theatrical. A 24-year-old fund manager built a huge public book around a single view of the future, then lost the ability to choose when to sell it.

These are reported private-fund figures, not audited public accounts. The fund's SEC Form 13F is useful but incomplete: it shows specified long US securities at a quarter-end snapshot, not shorts, private stakes, cash, many non-US holdings or the full leverage picture.

A forecast becomes a fund

Aschenbrenner's route to Wall Street ran through ideas first. Born in Germany, he finished Columbia College at 19 as the class of 2021 valedictorian, studying economics and mathematics-statistics. He helped organize the university's effective-altruism group, worked with the FTX Future Fund in 2022 and later joined OpenAI's Superalignment team. OpenAI dismissed him in April 2024. He said the firing followed his efforts to warn the board about security; the company said he had disclosed sensitive information improperly.

Two months later he published Situational Awareness: The Decade Ahead, a 165-page argument that AI capabilities would keep advancing rapidly, compute clusters would grow by orders of magnitude and power, chips and data centers would become strategic bottlenecks. “Trust the trendlines,” he wrote. The essay was prediction, policy memo and worldview. It also became an unusually detailed investment map.

Situational Awareness LP put money behind that map. Its disclosed public holdings leaned into the physical build-out: memory, cloud compute, power and data-center infrastructure. A first-quarter 2026 SEC filing listed 42 positions and roughly $13.7 billion in reported value, although option positions in 13F filings can make headline totals easy to misread. The private book included a large Anthropic stake. An investor letter reviewed by the Financial Times said the fund returned 439 percent net in the first half of 2026.

A portfolio can own dozens of tickers and still contain only one idea.

The hedge that repeated itself

A conventional long-short equity fund tries to separate stock-picking skill from the market's overall direction. It buys companies expected to outperform and shorts companies expected to lag. If the whole market falls, gains on the shorts may offset losses on the longs. The important work is not merely placing trades in opposite directions. It is making sure the risks underneath them are genuinely different.

Situational Awareness's public book looked balanced at first glance. The fund was long AI infrastructure names, including companies tied to memory, data centers and power. It was short software companies it believed AI could disrupt. Yet both legs asked the market to validate the same schedule. Fast AI adoption needed to raise the infrastructure winners and punish the software incumbents. A delay could do the reverse to both.

In July that shared dependency showed itself. AI-infrastructure stocks dropped sharply while software stocks rose. Longs lost money because they fell; shorts lost money because the borrowed shares rose. Bloomberg columnist Matt Levine described this kind of construction as rotating risk rather than removing it. Patrick Boyle supplied the cleaner joke in the reference video: it was like betting on one football team to win, then “hedging” by betting its opponent would lose.

Correlation often hides one level below a ticker symbol. Five companies can operate in different industries and countries, yet still respond to the same interest rate, commodity, customer or narrative. For investors, this is the part worth stealing: label positions by the condition that makes them work. If every label reads the same, the portfolio is concentrated no matter how many rows appear on the spreadsheet.

Leverage edits the ending

Concentration explains the direction of the loss. Leverage explains its speed. Reports said the fund had taken three to four dollars of exposure for each dollar of investor capital in parts of the book. Borrowing makes a winning return larger, but it also gives lenders a claim on the timing. When collateral falls below agreed levels, a prime broker does not have to admire a long-term forecast. It asks for cash, cuts exposure or sells.

This asymmetry is volatility drag in its simplest form. Suppose a coin flip wins 50 percent on heads and loses 40 percent on tails. The arithmetic average is a positive five percent per flip. But one win and one loss turn $100 into $90. Large swings reduce compound growth even when the average outcome looks favorable. Add leverage and expected return rises roughly in proportion to exposure, while variance rises with the square of it.

The formula is not the scandal. The deadline is. A leveraged investor may never reach the favorable future sitting in the far end of a probability distribution. One sufficiently bad path brings a margin call first. The forced seller realizes losses; a buyer with cash receives the option to wait. During the July unwind, Citadel beat other reported bidders for most of the public book. It was a transaction, not a charitable rescue.

Being early becomes indistinguishable from being wrong when someone else controls the clock.

The asset that could not run

An unlikely protection in the portfolio was illiquidity. Situational Awareness retained private holdings, notably a multi-billion-dollar stake in Anthropic, according to reporting after the sale. A private position has serious drawbacks: uncertain marks, limited buyers and long settlement. In this episode, those frictions also made it difficult for a lender to liquidate overnight. The part of the fund that could not be sold quickly helped it survive the part that had to be.

This does not make illiquidity a hedge. It reveals a second clock. Public markets mark continuously; private markets move by financing round, negotiated sale and valuation committee. Mixing the two can soften a visible drawdown while making the true portfolio harder to measure. Reports placed the slimmed-down fund between $8 billion and $10 billion. That estimate depends heavily on what the Anthropic stake is worth, a question no public screen can settle.

What survives the wreckage

Aschenbrenner's original argument has not been adjudicated by one month of market prices. Data centers are still being built. AI labs still demand more compute and power. Software may yet be transformed. Markets can disagree with the timing of a thesis without disproving its destination. That distinction is precisely why portfolio construction matters: a position must be able to survive the gap between an idea and its arrival.

His biography helps explain why the bet attracted Silicon Valley capital. He is young, technically informed, willing to make a legible forecast and prepared to stake a career on it. Those traits are rewarded in technology, where a concentrated founder can attempt something improbable and equity absorbs the failure. A hedge fund uses different plumbing. Debt has covenants. Shorts can rise without limit. Prime brokers call. The culture of the moonshot meets the mechanics of daily collateral.

There is also a portfolio lesson for the investors who backed him. Many were already economically long AI through jobs, founder equity and venture stakes. Adding a leveraged AI fund could compound the same exposure rather than diversify it. The clean question is personal: what happens to the rest of your life when this investment falls? If the answer is “the same thing,” the risk is larger than the statement shows.

The episode may leave Aschenbrenner with what he called expensive scars in a reported letter to investors. It leaves everyone else with a cheap checklist. Identify the common factor beneath every position. Size the trade for an ugly path, not an attractive average. Match the liquidity of assets to the liquidity of funding. And remember that conviction answers what you believe. Risk management answers whether you will still own it when the future arrives.