Leopold Aschenbrenner's hedge fund disclosed a $117 million U.S. equity portfolio in its maiden 13F filing, concentrated entirely in the infrastructure layer beneath artificial intelligence compute and Bitcoin mining. No positions in Anthropic's public competitors. No Microsoft. The bet is transformers, data centers, and electrical grid capacity—not the models themselves.
The filing shows stakes in publicly traded Bitcoin miners including Marathon Digital and Riot Platforms, power utility operators, and datacenter real estate trusts. Aschenbrenner, who left OpenAI in 2023 after authoring internal memos on compute scaling and superalignment timelines, launched the fund in late 2024. The portfolio construction suggests he believes the marginal dollar in AI goes to kilowatt-hours and rack space, not parameter count. His public essays have argued that models will saturate usefulness by 2027 and that the constraint shifts to physical infrastructure within eighteen months.
The positioning matters because Aschenbrenner carries credibility in both frontier labs and quantitative capital. He co-authored OpenAI's recursive reward modeling research and separately published a 50,000-word essay on AGI timelines that circulated among defense strategists and sovereign wealth allocators. If his fund is buying miners and utilities instead of NVIDIA or hyperscalers, the implication is that training runs have already crossed into diminishing returns and the next phase is operational deployment at territorial scale. That view runs counter to the consensus allocation into semiconductor and cloud infrastructure, which has driven $480 billion in market cap expansion since January 2024.
The 13F also reveals no exposure to the usual AI stack: no chip designers, no hyperscaler debt, no SaaS layers. The absence is as informative as the presence. Aschenbrenner's thesis appears to be that model capabilities have already cleared the threshold for economic utility and that the bottleneck is now energy arbitrage and datacenter logistics. Bitcoin miners fit this frame because they operate flexible compute loads with contracted power at sub-market rates, making them re-deployable for inference workloads. Several miners have already announced pivots to AI hosting, and their equity has traded sideways while hyperscalers have doubled.
Allocators should watch three follow-on signals. First, whether other former lab researchers file similar 13Fs in Q2 2025—if Aschenbrenner's positioning reflects insider knowledge of model plateau rather than idiosyncratic thesis. Second, whether Bitcoin miners announce datacenter partnership terms in the next 90 days, which would confirm the infrastructure-reuse arbitrage. Third, whether power utilities in Texas, Wyoming, and upstate New York see unusual volume in May and June, signaling that other funds are replicating the trade. The filing itself is retrospective to December 31, 2024, so the fund may have already rotated.
The portfolio is small enough to move quickly and large enough to reflect conviction. Aschenbrenner has not granted interviews since the filing.