Embattled hedge fund Situational Awareness invests $400M in chip startup Source Foundry

Leopold Aschenbrenner is twenty-six years old. He has never worked a trading desk. He launched Situational Awareness in 2024 on the strength of an OpenAI research résumé and a thesis that AI infrastructure would mint money. For a moment the market agreed. The fund swelled to twenty billion dollars. Then the AI infrastructure trade reversed. Aschenbrenner watched half his capital vanish in months. In July he sold the public book to Ken Griffin's Citadel — a white-flag surrender disguised as a transaction — retaining only the Anthropic position. The message was unambiguous: the liquid bets were wrong. The private conviction remains.

Now he has deployed four hundred million dollars into Source Foundry. That brings the fund's total exposure to the Stanford chip startup to half a billion. Half a billion from a ten billion dollar rump. Five percent of the entire fund in a single pre-revenue semiconductor venture. This is not portfolio management. This is a founder making a founder bet with other people's money.

Source Foundry promises faster cheaper chip manufacturing. The pitch sits at the intersection of two narratives that have captivated venture capital: AI compute hunger and American semiconductor reshoring. The Wall Street Journal broke the investment. The Journal did not report a valuation. That omission matters. At five hundred million committed Situational Awareness likely owns a controlling stake or something close to it. Aschenbrenner has effectively annexed a startup. He did not have the capital to do this three months ago. He does now because he liquidated everything else.

The Anthropic hold is the tell. Aschenbrenner kept the one asset that maps to his OpenAI provenance. He understands large language models. He understands the compute stack beneath them. He does not understand public market beta. The Citadel sale was an admission: I cannot trade this cycle. The Source Foundry bet is a repetition: but I can pick the next infrastructure winner. The logic is circular. The only evidence for his picking ability is the early returns that the drawdown just erased.

Ken Griffin bought the portfolio. Griffin runs Citadel with the discipline of a man who has survived multiple crises. He does not buy distressed books from twenty-six year olds unless the price embeds a wide margin of safety. Aschenbrenner took that price. He chose liquidity over conviction on the public side. Then he turned around and concentrated that liquidity into a single illiquid chip bet. The contradiction is stark. Either the public positions were never high conviction — in which case why were they the bulk of the fund — or the Source Foundry conviction is reckless overcompensation.

The wedding detail landed in the Journal story like a stray sentence. Aschenbrenner didn't let those setbacks get in the way of his wedding. The framing is generous. A more clinical reading: the fund manager attended his own wedding while his investors absorbed a fifty percent drawdown. The ceremony proceeded. The capital did not. That separation — personal milestone proceeding on schedule while professional stewardship fractures — is the profile of a man who experiences his fund as an extension of self rather than a fiduciary vessel. That profile produces the Source Foundry concentration. It also produces the Anthropic hold. It produces the Citadel sale. Every decision traces to the same psychology: I know what I know. Everything else is noise.

The investors who remain have no liquidity. The fund gates are likely shut. The mark-to-market on Source Foundry will be whatever Aschenbrenner says it is until a priced round arrives. The mark-to-market on Anthropic will be whatever the secondary market whispers. The public book is gone. The fee base has halved. The operating leverage on a ten billion dollar fund with a five hundred million single-name risk is existential. A fifty percent haircut on Source Foundry wipes two and a half percent of NAV. A zero writes five percent. The fund cannot absorb many zeros.

Aschenbrenner's thesis is that the next compute paradigm will be built not bought. Source Foundry is the build. Anthropic is the buy. The middle — the public semiconductor names the fund once owned — was the error. The correction is to double the build bet. The logic holds if the build succeeds. The logic collapses if the build stalls. There is no hedge. There is no diversification. There is a twenty-six year old's conviction financed by institutions that bought a hedge fund and received a venture fund.

The wedding proceeded. The fund contracted. The bet concentrated. The story is not over. But the shape of the ending is already visible in the structure of the position. Five hundred million dollars in one chip startup. Ten billion dollars total. One decision maker. No trading experience. No stop loss. The market will price the outcome. Aschenbrenner has already priced his conviction. The gap between those two prices is where his investors live now.