A hedge fund built entirely around the thesis that artificial intelligence would reshape markets faster than anyone expected has just been forced to confront how quickly that same thesis can unwind. NEWSCENTRAL sees this as one of the more vivid illustrations yet of how concentrated conviction bets on the AI trade can turn from a strength into a liability the moment sentiment reverses.
Situational Awareness, the fund founded by former OpenAI researcher Leopold Aschenbrenner, attempted to sell stakes in some of its privately held portfolio companies this week as it scrambled to raise cash against a wave of margin calls from its lenders, approaching venture firms Sequoia Capital and Greenoaks about taking over the positions before ultimately striking a separate arrangement with Citadel to offload the bulk of its public equity holdings instead.
Freddy Miller, Senior Analyst at NEWSCENTRAL, notes that the speed of the fund’s drawdown says more about leverage than about the underlying investment thesis. “Assets falling from $45 billion to roughly $10 billion in a matter of weeks is not simply a story about AI stocks correcting, it is a story about how much borrowed money was riding on those stocks staying at their previous valuations,” Miller notes.
The fund’s public portfolio had concentrated heavily in companies tied to the AI infrastructure buildout, including Bloom Energy, semiconductor maker Sandisk and cloud provider CoreWeave, all of which have been hit hard in recent weeks amid broader questions about whether AI-linked valuations can be sustained at their current levels.
Its private holdings tell a similar story of concentrated conviction: a stake in Anthropic worth more than $5 billion, alongside positions in infrastructure startups FluidStack and MatX, all bets placed on the assumption that foundational AI companies would keep compounding in value faster than public markets could reprice them.
Lucas Grant, Semiconductor and Manufacturing Strategy Analyst, underscores that the fund’s exposure to names like Sandisk and CoreWeave reflects a broader vulnerability across AI-adjacent portfolios right now. “Semiconductor and cloud-infrastructure names carry outsized weight in almost every AI-themed portfolio built over the past two years, so when sentiment turns, the drawdowns in those specific names tend to be sharper and faster than in the broader market,” Grant underscores.
We in NEWSCENTRAL warn that the fact Sequoia Capital and Greenoaks, both already investors in Anthropic, were the natural buyers approached for those private stakes shows how tightly concentrated ownership of the most prized AI assets has become among a small circle of overlapping investors.
That concentration cuts both ways: it gave Situational Awareness ready counterparties to approach when it needed liquidity quickly, but it also means that any further distress at similarly leveraged funds could pressure the same small set of private company valuations that Sequoia, Greenoaks and others are all holding simultaneously.
We at NEWS CENTRAL assess that this episode is unlikely to be an isolated one, since a fund built on aggressive conviction and leverage is exactly the kind of structure most exposed when a hot theme cools even briefly, and the AI trade has produced no shortage of similarly leveraged vehicles over the past two years.