Leopold Aschenbrenner, the former OpenAI safety researcher who left the company in May 2023, filed his first 13F quarterly disclosure for Situational Awareness LP, revealing a $13.7 billion portfolio split between concentrated long positions in energy infrastructure and notable put option exposure to semiconductor manufacturers. The filing landed without advance notice, marking Aschenbrenner's transition from policy commentary to capital allocation.
The portfolio structure reflects two distinct theses. Energy and infrastructure holdings comprise roughly 58% of disclosed long positions, concentrated in utilities with exposure to data center power contracts and transmission capacity buildout. The semiconductor puts—spread across six positions in the Philadelphia Semiconductor Index components—represent $2.1 billion in notional exposure at filing date, with strike prices clustered 12-18% below current trading levels and expiration dates concentrated in Q2 2025. The filing shows no corresponding call positions, suggesting directional conviction rather than spread construction.
This matters because Aschenbrenner spent eighteen months publishing detailed breakdowns of AI compute trajectories and power infrastructure bottlenecks through his Substack and podcast appearances. The portfolio reads as positioning for his stated view: that AI scaling will encounter power constraints before chip supply constraints, creating asymmetric upside for utilities with contracted capacity and asymmetric downside for semiconductor names trading at forward multiples that assume unconstrained demand growth. The $13.7 billion AUM figure places Situational Awareness among the top quartile of debut hedge fund disclosures in the past decade, indicating institutional backing aligned with a multi-year thesis rather than tactical momentum positioning.
The energy side shows specificity. Holdings tilt toward regulated utilities in PJM and ERCOT markets where hyperscale operators have announced power purchase agreements in the past eight quarters. Three of the top five long positions are utilities that reported contracted capacity increases exceeding 400 megawatts in their most recent earnings calls. This is not a broad energy thesis. It is infrastructure positioning ahead of a build cycle that Aschenbrenner has argued publicly will require $500 billion in transmission and generation investment by 2030 to support frontier AI training runs.
Operators and allocators should watch three events. First, whether Situational Awareness adds to semiconductor puts in the Q1 2025 filing, which would signal conviction that chip names have not yet priced in demand deceleration. Second, any equity raises or co-investment vehicles launched around the fund, which would clarify whether this is a personal portfolio or the foundation for a broader platform. Third, utilities earnings in late April and early May, when PJM and ERCOT operators will update contracted capacity figures and forward price curves for industrial load.
The filing came eleven months after Aschenbrenner's departure from OpenAI and seven months after his essay arguing that AGI timelines would compress power infrastructure rather than expand chip production. The $2.1 billion in semiconductor puts expire three weeks before Nvidia's May earnings call.