Leopold Aschenbrenner's debut 13F reveals $13.7 billion portfolio, heavy semiconductor shorts

Leopold Aschenbrenner, the former OpenAI safety researcher who published the widely-circulated "Situational Awareness" memo predicting AGI by 2027, filed his first 13F quarterly disclosure last week. The portfolio carries $13.7 billion in disclosed long positions and options contracts, an unusually large debut for a fund launched less than twelve months ago. The filing shows concentrated exposure to semiconductor manufacturers, significant put positions on chip stocks, and allocations to energy infrastructure—positioning that reads like a hedged expression of the memo's core thesis: compute will bottleneck AI, and energy will bottleneck compute.

The largest disclosed holdings include substantial long equity positions in NVIDIA, Taiwan Semiconductor, and ASML, alongside put contracts on those same names and on the broader semiconductor index. The fund also holds meaningful positions in Vistra Energy and Constellation Energy, two utilities with nuclear exposure. Total notional value of the options positions exceeds $4.2 billion, suggesting either aggressive hedging or directional bets on volatility in chip valuations. The filing does not break out strike prices or expirations, but the size of the put book implies Aschenbrenner expects either a material correction in semiconductor equities or significant realized volatility over the next twelve months.

What makes this filing noteworthy is not the size—$13.7 billion is substantial but not unprecedented for a well-capitalized launch—but the identity of the filer and the timing. Aschenbrenner left OpenAI in April 2023 after internal disputes over safety protocols, then spent six months drafting a 165-page memo arguing that AGI would arrive by 2027 and that national security depended on U.S. dominance in compute infrastructure. The memo circulated among defense contractors, semiconductor executives, and family offices. This 13F is the first public glimpse of how he is expressing those views with capital. The portfolio structure suggests he believes chip stocks have priced in demand growth but not the supply-side constraints—fabrication timelines, energy bottlenecks, export controls—that could compress margins or delay deployments.

Operators should monitor three follow-on signals. First, whether Aschenbrenner's fund files additional disclosures in offshore vehicles or private structures not captured by 13F rules—many large positions in commodities, private AI companies, or non-U.S. equities would fall outside this filing. Second, whether semiconductor volatility spikes in the next sixty days; if he sized the put book expecting near-term drawdowns, recent chip stock strength may force early adjustments. Third, filings from other ex-OpenAI researchers or safety-focused allocators who may have seeded or co-invested in the fund. The memo's distribution list was narrow but influential. If similar portfolios surface, it would confirm a coordinated view on compute scarcity as the dominant AI trade for 2025.

The energy allocation is the cleanest tell. Nuclear-exposed utilities do not appear in typical tech-focused portfolios unless the thesis is infrastructure, not software. Aschenbrenner is betting that the constraint is not model architecture or talent, but the physical capacity to power the data centers required to train frontier models. That view has not yet been priced into energy equities the way it has been priced into NVIDIA.