Leopold Aschenbrenner, the 25-year-old who left OpenAI in April 2024 after disputes over AGI timeline disclosure, filed his first 13F through Athena Capital Research showing $22 million in short positions against Nvidia and the VanEck Semiconductor ETF. The December quarter filing marks the operational launch of a fund seeded quietly in late 2024, roughly eight months after his departure from OpenAI's superalignment team.
The 13F shows put positions on NVDA and SMH as the fund's two largest disclosed holdings. Aschenbrenner holds no long semiconductor exposure in the filing. The positions were established between October and December 2024, a period when Nvidia traded between $120 and $145 and SMH consolidated near $240. The fund's AUM was not disclosed, but the $22 million notional represents a concentrated bet for a debut vehicle. Athena Capital Research is domiciled in San Francisco with Aschenbrenner listed as sole managing member.
The positioning directly contradicts consensus allocator views on accelerated AI capex. Aschenbrenner's public thesis, outlined in his June 2024 essay *Situational Awareness*, argues that training compute plateaus by 2027 while inference economics collapse faster than semiconductor ASPs can adjust. He expects hyperscalers to shift capex from Nvidia's H200 and Blackwell chips toward custom inference silicon and memory-dense architectures by late 2025. If correct, Nvidia's datacenter revenue—$30.8 billion in Q3 2024—would face margin compression before Street estimates of 2026. The short timing suggests Aschenbrenner expects re-rating within twelve months, likely tied to Microsoft or Meta guidance revisions in mid-2025 earnings calls.
The filing arrives as Nvidia's forward P/E sits at 28x on 2025 estimates, below the 35x average it held through H1 2024 but still pricing in 40% revenue growth. SMH, which weights Nvidia at 21%, Taiwan Semi at 13%, and Broadcom at 8%, has stalled near all-time highs since November. Aschenbrenner's bet requires either a demand shock or a margin reset. His OpenAI pedigree gives the position signaling weight with allocators who track researcher-turned-PM moves, particularly those fluent in training-to-inference capex phasing.
Watch for Athena's Q1 2025 13F in mid-May to see if Aschenbrenner adds exposure to ASML, Applied Materials, or Broadcom—chips adjacent to his inference thesis—or if he covers into any Nvidia sell-off below $110. Microsoft and Meta report January 29 and January 30; any deceleration in AI capex language would validate the timing. Separately, monitor whether Sequoia or Andreessen Horessen, both rumored early backers of Athena, file amended 13Fs showing GP commit or advisory stakes.
Aschenbrenner has not taken outside LP capital beyond a small friends-and-family allocation. The $22 million short is personal conviction, not institutional AUM. That makes it a view, not a crowded trade—yet.