The Situation Has Changed

There is an impressive number in the Situational Awareness LP investor letter. Through the end of June, Leopold Aschenbrenner's fund was up 439% net for 2026. Pretty nice.

The Financial Times reports that the fund has since approached existing investors and lenders for fresh capital following a sharp decline during July's AI stock rout, and that it offered some investors the opportunity to purchase assets from its portfolio. One person familiar with the discussions characterized them as ad hoc rather than a formal, coordinated capital raise. The precise July loss and the amount of capital sought have not been disclosed.

Read those two paragraphs together and the shape of the thing becomes clear. This is not a fund that was wrong. It is a fund that was extremely right, borrowed against being right, and then discovered what leverage does when the tape turns.

The AI stock rout of July 2026 impacted the fund

The arithmetic of a very good six months

Situational Awareness was up roughly 270% after fees through May, according to Wall Street Journal reporting in early June that also put assets under management above $20 billion. If the fund finished June up 439%, June alone contributed something in the order of a 45% gain.

The fund reportedly uses several major Wall Street banks as prime brokers, which matters, because on July 28 the FT also reported that Wall Street banks including Goldman Sachs and JPMorgan had begun telling hedge funds with concentrated exposure to the worst-hit sectors to post additional collateral to maintain existing borrowing levels. Market participants stressed that most of those calls fired automatically, under contractual risk provisions, rather than by anyone's decision. Goldman had already flagged that gross hedge fund leverage rose in the first five months of 2026 at the fastest cumulative pace since it began tracking the data in 2016.

What actually fell

July was not a drift. The Nasdaq 100 fell roughly 10% over the month. The Philadelphia Semiconductor Index has dropped about 25% since peaking around 14,655 on June 22, after a twelve-month rally of more than 130%.

Korea is the real story, and it is the one Aschenbrenner singled out. The Kospi closed at 5,663.24 on Wednesday, down 5.98% in a session after touching an intraday low of 5,262.77, against an intraday peak near 9,114 on June 21. That is a decline of roughly 38% from the high and roughly a third over the month, the index's worst month on record. Seven circuit breakers have tripped this year. Seoul has convened emergency financial meetings and moved to restrict single-stock leveraged ETFs and ETNs, the retail products that turned a chip correction into a liquidation cascade. Samsung Electronics and SK Hynix each fell more than 14% in a single session on July 28.

The catalysts arrived in order. Moonshot AI's Kimi K3 landed on July 17 and revived the DeepSeek shock. Reports of a Chinese breakthrough in chipmaking hit on Monday. Meta signaled it had surplus GPU capacity to sell rather than only buy. GPU hourly rental rates fell around 31% in three weeks. SK Hynix delayed HBM4 expansion in favor of DDR5.

Named positions took it hard. Situational Awareness's March disclosures included substantial exposure to Oracle and AMD, both down roughly 20% in July. Smaller AI-linked holdings fell further: Nebius, Bloom Energy, the memory complex including SanDisk, and Sharon AI, the Australian neocloud that listed on the Nasdaq in February under SHAZ and which announced a US$1.32 billion five-year compute contract on July 16 for its first AI factory, to be built in New Zealand.

One caveat worth holding onto. In the Q1 13F, the disclosed positions in Oracle and AMD were put options, not shares, part of $8.46 billion of notional put exposure against chipmakers. If those positions were still structured that way in July, a 20% decline in Oracle and AMD helped the fund rather than hurt it. Filings do not show what the book looked like by mid-July, and "exposure" in a 13F is a direction-agnostic word. The honest position is that we know the fund lost money and we do not know from public filings which legs did the damage.

The essay, the firing, the fund

Aschenbrenner is 25. German-born, Columbia valedictorian in 2021 at 19 with a degree in economics and mathematics-statistics, then a researcher on OpenAI's Superalignment team until April 2024, when OpenAI fired him over an alleged information security leak. He has consistently disputed that account, saying he was dismissed for raising safety concerns internally.

Weeks later came Situational Awareness: The Decade Ahead, 165 pages arguing AGI would arrive between 2027 and 2030, that the constraint would be physical rather than algorithmic, and that of the general public «few have the faintest glimmer of what is about to hit them». The buildout, he wrote, would be «a race to mobilize America's industrial might». Michael Dell shared it. So did Ivanka Trump.

Then he traded it. With AI researcher Carl Shulman he raised roughly $225 million in September 2024 from Stripe's Patrick and John Collison, Nat Friedman and Daniel Gross, set a $25 million minimum and a two-year lock-up, and pitched the vehicle explicitly against a venture structure: AGI would be too consequential to express in illiquid ten-year funds. The bet was never on models. It was on what models need. In his formulation, «It's not algorithms that are going to be the bottleneck». It's electrons.

It worked, spectacularly. Up 47% after fees in the first half of 2025, roughly 200% for the full year, 439% through June 2026. An Anthropic stake taken in February 2025 at just above a $60 billion valuation, against $965 billion in the company's most recent share sale, reportedly grew to about a fifth of assets. Jane Street came in as an investor. When a stake in T1 Energy surfaced in May, the shares rose 23% in a day.

He had no professional trading experience before launching. The latest filing shows four investment professionals and eight employees in total, running a book that has been reported above $20 billion. Read that as ruthless efficiency or as thin operational cover; it is at minimum an unusual ratio of leverage to headcount.

The infamous Situational Awareness website from 2024, source: Situational Awareness

The July 24 letter

Aschenbrenner is not retreating. In a July 24 investor letter he acknowledged the fund had "not been immune" to the volatility, singling out Asia, then framed the decline as one of the best buying opportunities since early 2025 and invited investors to add capital on August 1. He named a potential Anthropic IPO among possible second-half catalysts.

Inviting capital into a drawdown is exactly what a manager with conviction should do, and funds that raised in March 2020 look prescient. But "add money on August 1" and "we are approaching lenders" and "would you like to buy some of our positions" are three answers to the same question, which is how to reduce leverage without selling into a falling tape. Selling assets to your own investors is the most interesting of the three: it raises cash without printing a mark on the exchange, and it moves positions at negotiated prices between a manager and the people whose capital he manages. That is legal, occasionally sensible, and structurally awkward.

Naming an Anthropic IPO as a catalyst carries the same double edge. It is plausibly the single largest position in the fund, held privately, which means it does not mark down daily while the margin-financed public book does. A fund with roughly a fifth of its assets in an illiquid private mark and the rest financed at least partly with borrowed money has a liquidity mismatch that only becomes visible in a week like this one. The private valuation holds the reported number up. The collateral call arrives in cash.

What to watch

The Q2 13F, covering positions as of June 30, is due by August 14. It captures the book eight days after the peak, before most of the damage, but it will show whether the semiconductor puts were still on and how much leverage sat in the long book at the top.

Hyperscaler capex remains the variable that decides whether this was a valuation reset or a repricing. Microsoft held its outlook steady at roughly $190 billion for calendar 2026 on Wednesday night, alongside $90 billion of revenue, 43% constant-currency Azure growth and a $678 billion commercial backlog. Meta lifted the floor of its range to $130 billion and watched free cash flow fall 91%. Alphabet guided to about $200 billion over twelve months and posted its first negative quarterly free cash flow. Amazon reports Thursday night. Jensen Huang, asked this week, said there is «no AI bubble», which is both what he would say and possibly true.

The buildout is intact. As BNC argued in April when the first cracks appeared, the revenue underneath this cycle is real in a way the dot-com comparison never was, and the circular financing is also real. What broke in July was not the thesis. It was the financing of a position in the thesis.

Aschenbrenner may still be right about the decade. A fund up 439% at the end of June and making ad hoc calls to lenders four weeks later is a demonstration that being right about the decade and surviving the month are separately priced. Leverage is the instrument that decides which one you are graded on.

The essay warned that few people understood what was about to hit them. It did not say the author was exempt.