Leopold Aschenbrenner turned roughly $225 million into a hedge fund that peaked around $45 billion in under two years, posting a 439% net return through June. Then leverage of up to 4x met a brutal month in AI infrastructure stocks, and the whole public book went to Ken Griffin’s Citadel in a single pre-market block trade. On this week’s Bricks, Bucks & Bytes rundown, Dustin, Patric, and Martin argued the thesis didn’t fail, the leverage did – and asked whether the circular financing propping up the AI trade deserves the same scrutiny.
The LinkedIn commentary arrived before the facts did, which is what irritated Martin enough to open the episode with it. Thousands of people who have never managed outside money suddenly had firm opinions on how a 25-year-old should have run his multi-billion dollar portfolio.
Patric’s response was the right one: “Let’s revisit the facts so that we can also judge those that pass judgment.” So that’s what the crew did, and what we’ve verified against the reporting below.
What Actually Happened
The verified timeline, minus the commentary
Aschenbrenner left OpenAI in 2024 and launched Situational Awareness LP on the back of his widely read essay about the trajectory of AI compute. The fund started with roughly $225 million and grew to as much as $45 billion at its peak, riding concentrated positions in AI infrastructure names like SK Hynix, CoreWeave, Micron, SanDisk, and Nebius. An investor letter reported by the Financial Times showed the fund up 439% net through June 30.
Then July happened. AI infrastructure stocks sold off hard while some of the fund’s short positions in software moved against it, and with reported leverage running as high as 4x, the margin calls arrived fast. On July 30, the fund sold its entire public book, longs and shorts, in a single block trade before the market opened. CNBC’s David Faber broke the sale; the Wall Street Journal named Citadel as the buyer. What remains is a private portfolio anchored by an Anthropic stake worth around $5 billion.
One correction to the on-air discussion worth making: Patric quoted the fund at $24 billion under management, while Dustin remembered $40 to $45 billion. The reporting sides with Dustin – $45 billion at peak, roughly $20 billion in the public book by the time of the unwind.
A Fund Called Situational Awareness Missed Its Own
Great picks, zero margin for error
Nobody on the show disputed the picks. Dustin called it possibly “one of the greatest runs ever in portfolio history” for a two-year stint, and drew the line that matters: “This isn’t FTX or something like that.” No scandal, no fraud allegations, no vanished client money. The failure was structural, and Dustin named the irony directly – a fund called Situational Awareness lacked awareness of how much leverage it was carrying on assets that had already run enormously.
I don’t think he had alpha. I don’t think he had edge… he had a unique worldview and people rallied behind it.Patric Hellermann, General Partner, Foundamental
Patric’s broader complaint was about the discourse. “Can we just appreciate nuance online again?” Building a $45 billion fund from nothing in two years means convincing sophisticated backers, running real compliance infrastructure, and still making the picks. Had it gone to zero, the criticism would fit. It didn’t. As Dustin put it, the man will likely be a billionaire, made people money, and will probably raise again with a hard-won lesson about leverage priced in.
The mechanics of the endgame are less sentimental. Wall Street talks, a 4x-levered book is hard to hide, and forced sellers get bought at a discount. Dustin’s phrase for it: financial war. The buyer of last resort in that war tends to do rather well.
The money stories behind the machines
Funding, leverage, and the deals reshaping construction and AI infrastructure. Weekly, no fluff.
Join 3000+ ReadersThe House of Cards Question
Dustin’s SEC thought experiment
The conversation then widened from one fund’s leverage to the market’s. Dustin ran a thought experiment: if he chaired the SEC, he would move against circular financing in AI – the pattern where a chipmaker invests in a customer, and the customer sends the money back as chip purchases. Force the AI labs to real debt markets, with real interest rates and real repayment schedules, and suddenly they need real financial models rather than an endless loop of friendly capital.
“When you have circular financing, you are building a house of cards, and one card gets removed and the whole thing can collapse,” he argued. Patric took the other side: we live in a credit-based society, and you could in theory regulate volatility out of markets entirely, but you’d regulate growth out with it. His position was the lesser of two evils – accept that some people skim fees in the middle, because the alternative is building nothing. Martin added the historian’s note that every crisis produces regulation against the last crisis, never the next one.
The tie-back to Dustin’s opening topic was the best line of the episode: “We need people in charge with situational awareness on what creates systemic risk.” And for our readers, that risk has a street address. The same leverage logic now underwrites physical construction – Meta’s El Paso data center campus is close to 90% debt-financed through its BlackRock venture. A levered pipeline can unwind faster than any normal construction downturn.
| Item | Figure | Reported by |
|---|---|---|
| Launch capital (late 2024) | ~$225 million | FinanceFeeds / SpotGamma |
| Peak assets under management | ~$45 billion | CNBC |
| Net return through June 30 | 439% | FT investor letter, via SpotGamma |
| Reported leverage | Up to 4x | CNBC |
| Remaining private anchor | ~$5 billion Anthropic stake | CNBC / FinanceFeeds |
| Holdings after the sale | ~$10 billion | CNBC |
A 25-year-old former OpenAI researcher who graduated Columbia as valedictorian at 19, then launched Situational Awareness LP in 2024 after publishing a widely read essay on AI’s trajectory. His fund became one of the most watched vehicles in AI investing on the strength of its returns before the July unwind. (Source)
It depends entirely on when they got in. The fund was up 439% net through June 30, so early backers remained far ahead even after the losses. But the fund posted large losses in the weeks before the sale, and holdings fell from roughly $20 billion to around $10 billion after the Citadel trade, so capital that entered near the peak was hit hard. (Source)
Arrangements where money loops between connected parties – for example, a chipmaker investing in an AI company that then spends much of that capital buying the chipmaker’s products. Revenue and investment become intertwined, which flatters both sides’ numbers and, in Dustin DeVan’s argument, builds systemic fragility because removing one participant can cascade through the whole structure.
Because the AI buildout that fills construction backlogs runs on the same ingredient that sank this fund: leverage. Gigawatt data center campuses are increasingly financed with close to 90% debt through asset-manager ventures. If AI revenue disappoints and that debt reprices, project pipelines can thin out far faster than a conventional construction cycle would. (Source)
Related Articles
BlackRock Now Owns 80% of Meta’s $14 Billion AI Data Center
Big Tech Promised $650 Billion in Data Centers. Most of It Isn’t Being Built
Data Centers Are Keeping US Construction Afloat
