Leopold Aschenbrenner, the former OpenAI safety researcher who published the widely-circulated "Situational Awareness" memo predicting AGI by 2027, disclosed concentrated holdings in bitcoin mining operators and power infrastructure companies in his fund's first SEC 13F filing. The positions, reported for Q2 2025, show zero exposure to large-cap AI software or hyperscaler equity. Instead, the portfolio reflects a thesis on scarce compute capacity and the energy bottleneck underneath it.
The filing reveals long positions in at least three publicly traded bitcoin mining firms—names that derive revenue from selling both mining capacity and physical electricity access to third parties. Aschenbrenner's fund also holds positions in power generation and grid infrastructure companies with exposure to data center build-outs in Texas and the Mountain West. Combined, these positions represent more than 60% of disclosed assets under management, per the filing. The fund launched in late Q1 2025 with an initial pool believed to be in the low nine figures, sourced from a mix of single-family offices and venture-adjacent LPs who backed Aschenbrenner's departure from OpenAI in mid-2023.
The allocation carries a clear second-order thesis. If AI compute demand continues to double every six months—as Aschenbrenner argued in his September 2024 memo—then power access becomes the binding constraint before chips do. Bitcoin miners sit on gigawatt-scale power contracts negotiated years ago, often in jurisdictions with stranded renewable capacity. Several miners have already pivoted to leasing that capacity to hyperscalers for AI inference at margins 300-500 basis points above bitcoin mining economics. The fund's position is a pure-play bet that this lease arbitrage accelerates as Microsoft, Meta, and Google exhaust available grid capacity in traditional data center markets.
What makes the positioning unusual is Aschenbrenner's public profile. He spent eighteen months inside OpenAI's superalignment team, left under unclear circumstances, and has since become the most vocal proponent of a hard-takeoff AGI timeline among credible technical voices. His investor letters reportedly include detailed models of training run costs, chip allocation, and power draw per parameter—technical scaffolding rare in equity hedge funds. The bitcoin miner bet suggests he believes the infrastructure gap is two to four quarters ahead of where the market currently prices it. If correct, the stocks he owns would re-rate as power infrastructure plays, not speculative crypto proxies.
Operators and allocators should watch three follow-on signals. First, whether any of the named miners announce capacity lease agreements with hyperscalers in Q3 or Q4 earnings calls—those deals typically carry 18-24 month lead times and would validate the thesis. Second, whether Aschenbrenner's fund discloses additional positions in private power developers or behind-the-meter generation assets in future filings or LP letters. Third, whether venture-stage AI labs begin bidding directly for power contracts independent of cloud providers, a shift that would further tighten available capacity. That last signal would likely surface in permitting filings in Texas, Montana, and Wyoming within the next six months.
The filing lands three weeks after Aschenbrenner published a follow-up essay arguing that the U.S. has 12-18 months to secure domestic compute advantage before Chinese labs close the gap. His fund's equity book now suggests he is positioning for the constraint that arrives first.