Leopold Aschenbrenner's Fund Unwinds After 439% Peak Return

From Superalignment to Superleverage

Situational Awareness, the hedge fund launched by former OpenAI researcher Leopold Aschenbrenner in 2024, has offloaded the majority of its publicly traded equity positions to Ken Griffin's Citadel after a sharp drawdown erased months of outsize gains. The move marks a dramatic reversal for a 25-year-old who entered the year as one of venture capital's most closely watched figures, riding a thesis that AI's exponential scaling would require unprecedented investment in chips, data centers, and power infrastructure.

The fund returned 439% through June, according to figures disclosed to investors. Assets under management reportedly reached $45 billion at their peak before a sector-wide repricing of AI infrastructure names triggered a cascade of losses. By the time Citadel stepped in to acquire the bulk of those positions, Situational Awareness' total assets had contracted to roughly $10 billion, down from approximately $20 billion in recent months.

The German-born Aschenbrenner, who enrolled at Columbia University at 15 and graduated valedictorian at 19, joined OpenAI's superalignment team in 2023. He was dismissed a year later over what the company described as an improper disclosure of internal information. That team, then led by co-founder Ilya Sutskever and researcher Jan Leike, has since disbanded; Sutskever founded his own venture, while Leike moved to Anthropic. Aschenbrenner launched Situational Awareness shortly after his departure, raising several hundred million dollars from backers including Jane Street, Stripe co-founders Patrick and John Collison, and former Meta executives Daniel Gross and Nat Friedman.

Leverage Amplifies the Drawdown

The fund's strategy centered on a concentrated portfolio of AI infrastructure equities: memory chip producers SK Hynix and Sandisk, clean-energy developer Bloom Energy, and neocloud provider Nebius Group. All four names have fallen more than 30% over the past month as public-market investors questioned whether hyperscale capital expenditures would translate into near-term revenue growth. At DailyTechWire, we've tracked similar skepticism ripple through earnings calls at Microsoft, Amazon Web Services, and Google Cloud, where executives have struggled to articulate clear payback timelines for multibillion-dollar data-center commitments.

Situational Awareness amplified its exposure through leverage, a common hedge-fund technique of borrowing against existing positions to buy additional shares. When prices move in the anticipated direction, leverage magnifies returns; when they reverse, losses compound quickly. In a July 24 letter to investors, Aschenbrenner characterized the sell-off as one of the best buying opportunities since early last year and invited clients to commit fresh capital starting August 1. The appeal did not generate the inflows he anticipated, forcing a negotiated exit with Citadel.

Griffin's firm has a well-established pattern of acquiring distressed positions from overleveraged counterparties. Citadel's own portfolio already included several of the same AI infrastructure names, signaling that the firm shares Aschenbrenner's long-term view but possesses the balance-sheet capacity to weather short-term volatility. For a fund with only two years of operating history and no prior trading experience among its principals, that cushion proved unavailable.

The Anthropic Wildcard

Situational Awareness did not sell its private holdings, the most significant of which is a stake in Anthropic currently valued at $5 billion. The AI safety and research company was last valued at $965 billion in a Series H round completed in May and is widely expected to pursue a public listing as soon as October. If Anthropic prices its initial offering at or above that private valuation, the resulting gain could offset a substantial portion of the fund's public-market losses.

Anthropic's trajectory mirrors the broader divergence between public and private AI valuations. While listed semiconductor and infrastructure stocks have faced pressure from earnings misses and margin compression, private AI labs continue to attract capital at escalating prices. Investors in those companies are betting on longer development cycles and the potential for winner-take-most dynamics in foundation models, reasoning that tolerates deferred revenue in exchange for strategic positioning.

Beyond Anthropic, Situational Awareness holds stakes in chipmaker MatX and AI data-center startup Fluidstack, which was reportedly in talks in April to raise a new round at an $18 billion valuation. Both companies remain pre-revenue or early-revenue, meaning their valuations rest almost entirely on projected demand for specialized inference hardware and edge compute capacity. If that demand materializes as Aschenbrenner originally forecast, the private book could validate the investment thesis even as the public book faltered.

Timing, Temperament, and the Cost of Conviction

The collapse of Situational Awareness' public portfolio offers a case study in the risks of conviction investing without the institutional scaffolding to support it. Aschenbrenner's essays on AI scaling, published before the fund's launch, laid out a coherent argument: training runs would grow by orders of magnitude, requiring commensurate investment in memory bandwidth, interconnect fabric, and power generation. The thesis was not wrong in direction, but it proved vulnerable to the timing mismatch between capital deployment and revenue realization that has bedeviled infrastructure investors throughout 2026.

Hedge funds with longer track records and deeper benches typically manage concentration risk through position limits, stress testing, and dynamic deleveraging protocols. Situational Awareness, by contrast, appears to have maintained high conviction and high leverage simultaneously, a combination that leaves little room for error when correlations spike and liquidity evaporates. The July letter to investors, sent even as positions were hemorrhaging value, suggests a temperament more suited to venture capital, where marks adjust slowly and patience is rewarded, than to public equities, where margin calls arrive overnight.

Ken Griffin's willingness to step in reflects a different calculus. Citadel can afford to hold these names through multiple earnings cycles, capturing any recovery in sentiment without the pressure of investor redemptions or financing covenants. For Aschenbrenner, the question now is whether the Anthropic stake and other private positions appreciate quickly enough to rebuild credibility and capital. An October IPO at a premium valuation would provide both, along with a narrative of vindication. A delay or a down round would make the public unwind look less like tactical repositioning and more like a miscalculation of market structure and risk capacity.

The coming months will determine whether Situational Awareness re-emerges as a differentiated vehicle for AI infrastructure investing or becomes a cautionary tale about the perils of leverage, concentration, and youth in a sector where the gap between thesis and execution has never been wider.