The Manifesto Fund: How Leopold Aschenbrenner Built The Fastest-Growing AI Hedge Fund in the Country
- David Borish
- 14 minutes ago
- 7 min read
From Firing to Fund
I first wrote about Leopold Aschenbrenner when he was fired from OpenAI in April 2024. The official reason was leaking internal information. Aschenbrenner said the disputed document was a safety brainstorming memo he shared with three external researchers for feedback, and that an earlier memo criticizing OpenAI's security posture against foreign actors was the real cause. Two months after being dismissed, he published Situational Awareness: The Decade Ahead online. Ivanka Trump shared it. It circulated in Silicon Valley, policy circles, and AI labs. Within months, Aschenbrenner had raised anchor capital and launched Situational Awareness LP.
The fund's early backers were not traditional finance figures. Patrick and John Collison, co-founders of Stripe, came in early. So did Nat Friedman, the former GitHub CEO who now leads AI products at Meta, and Daniel Gross, Friedman's investing partner who now co-leads Meta Compute. One anchor LP told Fortune that Aschenbrenner had explicitly rejected a venture-capital structure when pitching the fund. His argument was that if AGI arrives on the timeline he forecasted, the right way to capitalize on it is through public markets, which reprice faster and scale with conviction. Private startup bets would be too slow and too narrow.
The fund launched in September 2024 with its first 13F filing, covering Q4 2024, disclosing roughly $254.8 million in six holdings. By the time Fortune profiled Aschenbrenner in October 2025, the fund had grown to over $1.5 billion in assets under management and had delivered 47% returns in the first half of the year, after fees. An index of tech hedge funds compiled by PivotalPath returned 7% in the same period. The S&P 500 returned 6%.
Aschenbrenner had, in the span of roughly a year, gone from being fired to running one of the fastest-growing new hedge funds in the country, and doing so with no prior professional investing experience.
What the 13F Filings Show
Situational Awareness LP files quarterly 13F reports with the SEC, which require disclosure of long positions in U.S.-listed securities. These filings provide a partial but telling picture. Shorts, derivatives, international investments, and private stakes are not disclosed, so the full book is larger and more complex than what appears in the public record.
What the filings do show is a portfolio built almost entirely around one thesis: the physical infrastructure required to run AI at scale is underpriced relative to where demand is heading.
The Q4 2024 filing showed six positions totaling roughly $255 million. By Q1 2025, the fund had added 20.2 million call options in Intel. By Q3 2025, the disclosed positions had grown to approximately $4.1 billion. The Q4 2025 filing, submitted February 11, 2026, disclosed 29 holdings worth $5.52 billion.
The portfolio's top ten positions account for roughly 86% of the book, a concentration level unusual even by hedge fund standards. The largest disclosed position as of Q4 2025 is Bloom Energy, a company that manufactures solid-oxide fuel cells for on-site power generation at data centers, with 10.1 million shares and 408,500 call options worth approximately $875 million. Bloom's stock rose an estimated 176% from the fund's entry point in Q4 2025 through early May 2026.
The second-largest category involves Bitcoin miners that have been repurposing their power-dense facilities as AI computing hosts. Core Scientific, IREN, Cipher Mining, Riot Platforms, Bitdeer, and Applied Digital all appear in the filings. The logic is that these companies sit on exactly the assets the AI buildout needs: cheap land, access to power, and existing high-density computing infrastructure. After the 2024 Bitcoin halving squeezed mining margins, several of them had already begun converting their operations. Core Scientific signed a series of 12-year high-performance computing contracts with CoreWeave. IREN and Applied Digital have pursued similar pivots.
Earlier filings showed large positions in traditional semiconductor plays. During Q1 2025, the fund held substantial Intel call options, a bet that looked contrarian at the time. Intel's shares later climbed following a federal investment and a subsequent $5 billion stake from Nvidia, validating at least the timing of the position. The fund simultaneously held short exposure to the broader semiconductor sector through other instruments, according to reporting from Fortune, while remaining long select names. The coexistence of Intel longs and semiconductor shorts reflects a more granular view: Aschenbrenner was not simply betting on chips as a category, but on specific bottlenecks within a supply chain he believed the market was mispricing.
The Thesis in Practice
The intellectual foundation for all of this is the 165-page essay, which argues that AI capabilities are improving on an exponential curve and that the compounding effects of compute, algorithmic efficiency, and what Aschenbrenner calls "unhobbling" — removing artificial constraints on what models can do — will bring AGI by approximately 2027 and superintelligence shortly after.
From that forecast, the investment logic follows directly. If AGI requires training clusters and inference infrastructure at scales the current grid cannot support, then the binding constraints are electricity generation, transmission, and computing capacity. The companies positioned at those bottlenecks are undervalued by investors who either don't believe the timeline or haven't worked through what it implies for demand.
The fund's public record suggests Aschenbrenner bought Bloom Energy, power producers Vistra and Constellation Energy, and eventually the Bitcoin miners before any of those categories became consensus AI infrastructure plays. He was also buying during moments when the consensus was going the other way. When DeepSeek's January 2025 release triggered a broad sell-off in AI infrastructure stocks, the Q1 2025 13F filings later showed the fund had been adding positions.
One notable absence from the portfolio is the companies building AI models themselves. There are no Nvidia shares, no Microsoft, no Google, no Meta. The only notable private-market exposure that has been reported is a position in Anthropic. The reasoning, consistent with the essay, is that the model builders are already well-understood and well-priced. The infrastructure required to run their models is not.
The Numbers Between the Raises
The period between the fund's September 2024 launch and the Fortune profile in October 2025 is where the growth story gets specific. The fund started with approximately $225 to $255 million in disclosed positions and grew to $1.5 billion in AUM within roughly a year. That growth came from two sources: capital inflows and performance.
The 47% return in the first half of 2025, after fees, confirmed that performance was doing significant work. Exact figures for how much of the fund's growth came from new LP capital versus appreciation have not been publicly disclosed. But the Q3 2025 13F, showing roughly $4.1 billion in disclosed securities, indicates the positions were growing substantially faster than any ordinary inflow rate would explain. Multiple sources indicate that top holdings generated returns of 100% to 800% from entry to peak over the 2025 period.
According to Fortune, Aschenbrenner has invested nearly all of his own net worth in the fund. At the Q1 2025 filing period, before the Intel and Bloom positions had fully appreciated, the fund's disclosed book was already in the hundreds of millions. Given the reported performance and his personal stake, the period between the first raise and the fund's October 2025 prominence was likely when the most concentrated personal wealth creation occurred, though specific figures remain private.
The fund's LP base had also broadened considerably by the time of the October 2025 Fortune profile. Beyond the anchor Silicon Valley names, the fund had taken on West Coast founders, family offices, institutions, and endowments. Many investors agreed to multi-year lockups, a structure that signals confidence and reduces redemption risk.
The Debates Around the Record
The legitimate questions about Situational Awareness LP are worth noting directly. The 13F filings capture only long U.S. equities and options, which means the full portfolio, including shorts and international positions, is not visible to the public. A fund that is short certain sectors while long infrastructure names would have a different risk profile than the filings alone suggest.
Several former OpenAI colleagues, according to Fortune's reporting, have questioned whether the fund's early results reflect skill or favorable timing. The AI infrastructure trade was not contrarian by the end of 2025 it had become a mainstream institutional position. The argument that Aschenbrenner identified it before the crowd is supported by the Q4 2024 filing, which predates the broad market recognition of power scarcity as an AI constraint, but skeptics note that getting the direction right in a strong macro tailwind does not by itself prove a repeatable analytical edge.
There is also the question of leverage. Reporting on the fund has noted an implied leverage ratio of approximately 14:1, based on options positions relative to stated equity exposure. If the AI capex cycle slows, if AGI timelines slip materially, or if a broader tech correction arrives, a portfolio this concentrated and this levered would face significant drawdown risk.
Aschenbrenner acknowledged this structure implicitly in the original essay. The entire thesis is a high-conviction, concentrated bet on a specific timeline. If the timeline is wrong, the portfolio is wrong. The fund is not hedged against the possibility that AGI takes longer than expected. It is, by design, long the certainty of it.
What the Portfolio Says Now
The February 2026 filing is the most complete picture of where the thesis currently sits. With 29 holdings and $5.52 billion in disclosed U.S. equity exposure, the portfolio has rotated substantially from its early semiconductor weighting toward power and energy infrastructure. Bloom Energy is the largest position. The Bitcoin miner basket is among the most recently added. SanDisk, Coherent, and Tower Semiconductor represent bets on storage, optical networking, and specialty chip fabrication — the quieter components of an AI data center that don't generate headlines but constrain throughput.
The portfolio is still missing the household names. That is either the fund's most distinctive feature or its most significant risk, depending on how the next few years unfold.
By any conventional measure, the track record through 2025 was hard to argue with. A 47% first-half return, a 22-fold increase in disclosed equity exposure over 12 months, and a growing institutional LP base are not outcomes that get explained away by luck alone, even if luck played a role.
The more significant question is whether a thesis designed for a specific two-to-three year window will remain the right portfolio for whatever comes after it.
Aschenbrenner is 24. He has, according to his own account in the original essay, invested essentially everything he has in the answer.
Because Situational Awareness LP files public 13F disclosures with the SEC every quarter, it's possible to see exactly what their holding. I built a simple tool that lets you mirror their disclosed positions at whatever dollar amount you want to put in. Enter a number, and it scales every position to match his weightings. A few caveats worth noting: 13F filings are delayed by up to 45 days, options positions aren't practical to replicate at most retail sizes (his largest single position is a CRWV call stack that represents 18% of the fund), and this is not investment advice. It's a window into how one of the more interesting minds in AI is actually positioning his money.
Comments