The prophet of
In June 2024 Aschenbrenner published Situational Awareness: The Decade Ahead, a sprawling essay running to about 165 pages. Its central claims were that artificial general intelligence arrives around 2027, that the training clusters required to get there would cost over $1 trillion by 2030, and that they would draw over 100 GW of power. He restated the argument at length on Dwarkesh Patel's podcast, where the trillion-dollar cluster and Chinese espionage at AI labs were the two recurring themes.
The forecast has not been vindicated in its specifics and has been comprehensively vindicated in its direction. Nobody has announced a $1 trillion cluster or a 100 GW single site. But one 2026 tally put combined planned capex for Amazon, Google, Meta and Microsoft at up to $630 billion in 2026, against $388 billion in 2025 for the same four; J.P. Morgan put the five largest US at $697 billion for 2026, a figure that had risen $173 billion since the start of that year. Power deals moved from megawatts to gigawatts on the same schedule: Energy Vault announced a 1.25 GW off-grid power infrastructure agreement for a Texas hyperscaler AI data centre on 7 August 2026, and NRG was reported on 6 August 2026 to be nearing a hyperscaler deal with a 1.2 GW initial phase.
Before any of this, Aschenbrenner was a fast-track product of the effective-altruism funding machine. He entered Columbia at 15, graduated as valedictorian in 2021 at 19, worked at the FTX Future Fund, and joined OpenAI's superalignment team in 2023. The Future Fund itself ceased operating after FTX's November 2022 bankruptcy.
The OpenAI dismissal
OpenAI fired him in April 2024, and the two accounts of why have never been reconciled. OpenAI said he had improperly shared internal information — specifically, a "brainstorming" document sent to three outside researchers for feedback. Aschenbrenner says the document had been scrubbed of sensitive material, that sharing such things externally was normal practice, and that the real reason was a memo he had sent the board warning that OpenAI's security was not strong enough to protect model weights and algorithmic secrets against foreign-state and Chinese industrial espionage. By his account he wrote the memo, was warned by HR, and was later dismissed on a pretext. OpenAI, per reporting, denied that his security concerns led to his separation.
Both stories can be partly true and neither is checkable from outside. What is checkable is what happened next: the man who had just argued in public that the frontier labs were sitting on the most valuable secrets in the economy went and set up a fund to trade the consequence.
The fund launched in 2024, when Aschenbrenner was 22, with about $225 million. The named backers in the reporting are and PIMCO, with one account referring to additional "well-known financiers" it did not name. That investor list is the single most important fact on this page.
The portfolio is unusually legible for a private fund, because US long equity and options positions have to be disclosed quarterly on a 13F, and the shape of the book changed violently. As of 31 March 2026 the fund reported $13.677 billion across 42 entries, of which $8.459 billion — about 62% — was put options, meaning the disclosed book was mostly short semiconductor and AI exposure. Three months later the Q2 2026 filing showed $20.242 billion across just 26 positions as of 30 June 2026, dominated by long memory: SanDisk at $5.674 billion (28.0%) and Micron at $5.574 billion (27.5%), together more than half the book, then Taiwan Semiconductor ($1.27 billion, 6.3%), Bloom Energy ($1.233 billion, 6.1%) and ($1.233 billion, about 6.1%). Separately, 2026 profiles reported a stake in Anthropic put at around $5 billion.
Regulatory filings tell a third story again. A Form ADV filed 27 April 2026 reported $9.28 billion of discretionary regulatory assets under management as of 31 December 2025. Reported performance since launch has been quoted at about 439% net through June 2026 in one account and roughly 270% after fees in 2026 through May in another. Nobody outside the firm has published the actual fee terms; where media pieces cite "2 and 20", they are using a generic hedge-fund model as an illustration, not his contract.
What none of the disclosures show is the . A 13F reports positions, not borrowings. The gap between a $20.2 billion disclosed long book and a fund reported at $45 billion is where the interesting part lived, and it was invisible until it wasn't.
The $35 billion margin call
In July 2026 the trade went the other way. CNBC reported that the fund collapsed from about $45 billion to around $10 billion after a rout in semiconductors, with the public equity portfolio sold to Citadel in a forced transaction. The Wall Street Journal reported the Citadel purchase of the stock book following the losses. From $225 million at launch to a reported $45 billion and back down by roughly $35 billion in a month, in two years.
The peak number is contested. Some July 2026 reporting put the fund at $20 billion to $24 billion rather than $45 billion, which is a wide enough range that we would not build an argument on either end of it. Estimates of Aschenbrenner's own paper wealth are worse: one 2026 analysis suggests it may have touched $5 billion at the peak and fallen to about $300 million; another post-collapse profile puts it nearer $100 million. Outlets do not even agree on his age at the time, describing him as 24 or, in a profile giving his birth year as 2001, 25. Private funds do not have to tell you these things, and the speed of the unwind meant that by the time anyone asked, the answer had changed.
He did not stop. On 6 August 2026 the New York Post reported that he had made a new $400 million bet after the near-collapse of the fund.
Why it matters to this crash
A man in his mid-twenties losing an enormous amount of money on a concentrated bet is a story. It is not a systemic story. What makes this one systemic is the counterparty.
Jane Street lost $15 billion in July 2026, its first losing month in a decade, on including its investment in Situational Awareness. The FT, ranking the event against history's biggest trading losses, put a $35 billion July figure at Aschenbrenner's leveraged AI stock bets. We wrote on 19 August that we cannot reconcile the two numbers from anything published — one may be a 's gross damage and the other a firm's net monthly P&L — and we are not going to pretend otherwise. The order of magnitude is not in dispute.
Jane Street is not a hedge fund investor in the way a pension is a hedge fund investor. It is a core liquidity provider in ETFs and options: its balance sheet is what stands between a redemption order and a price. In its most recent debt offering it described its own strategy as having evolved to include longer-term positions, in line with a hedge fund. So the entity you need to make prices in a selloff is the entity taking the losses in the selloff. A bank running that overlap would face capital charges and a supervisor. Jane Street is not a bank, funds itself in the , and discloses almost nothing, which is why nobody saw the overlap until a month like July.
The scale of that layer is the point. Jane Street still generated more than $40 billion of net trading revenue over the year to mid-August 2026. Hudson River Trading posted $11.4 billion of trading revenue and $7.4 billion of net profit in the three months to June 2026, a record. In 2025 Jane Street, Hudson River and Citadel Securities together took in more than $60 billion, beating the trading desks at JPMorgan and Goldman Sachs. We called it on 20 August: four are now the market, and they are unregulated, undisclosed, and demonstrably willing to lose fifteen billion dollars in a month on a directional bet.
Then there is the aftershock we found most unsettling. Banks carrying exposure to leveraged AI vehicles and to Situational Awareness through the summer had hedged the jump risk by buying semiconductor options. Once the exposure vanished, so did the hedges. Implied volatility on the VanEck Semiconductor ETF fell from 65% in July 2026 to 40%, the lowest since February, and the ratio of open puts to calls dropped to 1.89, the most call-heavy since early April. "Now they don't need those hedges, and I believe unwinding of those hedges has made volatility in the sector inexpensive," Zed Francis of Convexitas told CNBC. As we put it on 20 August: the failure did not make the system more cautious, it made caution cheap.
What would make this dangerous
The collapse of one fund financed by one prop firm was absorbed. Four things would turn it into something else.
A second one. Situational Awareness was funded by Jane Street and PIMCO and nobody outside those firms knows how many comparable vehicles exist, how levered they are, or which market makers are behind them. A 13F shows US long equity and listed options and nothing else. If a second fund of this shape unwinds into a market where hedges have already been sold — semiconductor implied vol at 40%, put/call at 1.89 as of mid-August 2026 — the buffer is thinner than it was in July.
A liquidity provider that steps back. The question after a $15 billion month is not whether the firm survived; it plainly did. It is whether a firm that has just taken a loss of that size on its own book still wants to warehouse everyone else's risk in the next stress. Watch bid-ask and ETF premiums in the next volatility event, not the P&L press releases.
Capex that turns out to have been funded this way. J.P. Morgan's $697 billion 2026 hyperscaler estimate for the is not obviously dependent on levered equity funds. But the marginal financing — the Blooms and Nebiuses of the world, the multi-gigawatt power deals signed in August 2026 — is where non-bank credit shows up first. If the appetite that funded Situational Awareness was also funding the buildout's second tier, the freeze arrives there before it arrives at Microsoft.
And the disclosure gap staying open. The BIS reported that non-bank financial institutions now account for more than 40% of non-centrally cleared euro-denominated cash-borrowing , up from under 30% at end-2020. Federal Reserve research puts the Treasury at roughly $830 billion as of September 2025, about double its 2020 peak. The migration of leverage out of supervised balance sheets and into private ones is the running thesis of this site. Aschenbrenner is the version of it with a face.