Leopold Aschenbrenner filed his first 13F as manager of Athena Capital with short positions against Nvidia and Oracle, disclosed holdings show. The fund reported $25 million in regulatory assets under management with put options expiring between March and June 2025. Aschenbrenner left OpenAI in April 2023 after publishing internal concerns about security protocols for superintelligent systems.
The filing lists 4,000 puts on Nvidia at strikes between $120 and $140, and 2,500 puts on Oracle at $150 and $165. Both positions sized at roughly 40% of disclosed book value. Athena also holds smaller bearish bets against Microsoft and Palantir, though call spreads on Anthropic-adjacent names suggest conviction in diffused AI infrastructure rather than concentrated hyperscaler dominance. The Oracle position stands out. The company trades at 28x forward earnings on database licensing that renewal data shows decelerating in enterprise accounts above $10 million annual contract value.
Aschenbrenner's public thesis centers on compute bottlenecks derailing the current AI investment cycle. His June essay *Situational Awareness* argued that training runs would require 100 gigawatts of power by 2028, a figure that implies either dramatic efficiency gains in chip architecture or a repricing of datacenter capex assumptions. Nvidia closed Friday at $131, down 18% from October highs, but still trading at 33x next year's earnings on analyst estimates that bake in 40% revenue growth. If Aschenbrenner is correct that the marginal cost curve for frontier model training steepens faster than hyperscaler budgets expand, then current multiples assume a linearity that disappears in 2025. The Oracle short makes sense in that frame. Cloud database migration stalls when enterprises realize the next marginal improvement in model capability requires rearchitecting data pipelines they just finished rearchitecting.
The 13F structure means Athena's gross exposure likely exceeds the $25 million reported. Funds disclose long positions and certain equity derivatives but omit credit instruments, private holdings, and non-equity hedges. If the put positions represent 20-30% of total AUM, Athena runs closer to $100 million, consistent with a seed-stage quantitative hedge fund backed by high-conviction LPs who read the same OpenAI internal memos Aschenbrenner did. The timing matters. Nvidia reports fiscal Q4 earnings on February 26. Oracle reports Q3 on March 10. If either company guides below Street expectations on datacenter utilization rates or shifts capex deployment timelines outward, Aschenbrenner's puts move in the money before March expiry.
Watch for follow-on filings from other ex-OpenAI researchers who left between April 2023 and June 2024. Seven senior researchers departed during that window, several citing misalignment between safety timelines and product shipping pressure. If two or three launch vehicles with similar disclosed positioning, the market reads it as insider information arbitrage, not idiosyncratic macro view. Also watch Nvidia's January 6 CES keynote. If Jensen Huang announces Blackwell Ultra delays or reprices B200 cluster configurations above $3 million per rack, the February earnings call becomes a volatility event and March puts reprice accordingly.
Aschenbrenner filed the 13F on December 13, meaning positions reflect November 30 marks. Nvidia closed that day at $135. Oracle at $162. Both have since drifted lower. The question is not whether Aschenbrenner called the top—it is whether the top he called is a 15% correction or the start of a 40% unwind when 2025 capex budgets hit physical infrastructure constraints that no amount of software optimization resolves.