Leopold Aschenbrenner Investment Partners disclosed short positions worth approximately $47 million against NVIDIA Corporation and the VanEck Semiconductor ETF in its inaugural 13F filing covering Q4 2024. The VanEck SMH short ranks first in the new fund's disclosed holdings, with NVIDIA puts occupying the second slot. Aschenbrenner, who left OpenAI in 2023 after authoring internal memos warning of AGI timeline compression, named his vehicle Situational Awareness after his widely-circulated essay predicting accelerated AI development would create severe resource bottlenecks.
The positioning represents a directional bet against the thesis Aschenbrenner himself promoted in public writing. His June 2024 essay "Situational Awareness: The Decade Ahead" argued that compute demand would exceed supply through 2027, requiring $100 billion in annual capital expenditure from hyperscalers. The 13F short suggests Aschenbrenner now sees semiconductor pricing or demand inflection arriving sooner than consensus expects. The filing shows no offsetting long positions in alternative chip architectures or cloud infrastructure, indicating a clean duration short rather than a pairs trade.
The timing matters for three reasons. First, NVIDIA reported January earnings showing $22.1 billion in datacenter revenue, a 409% year-over-year increase, with management guiding Q1 revenue to $24 billion. Second, SMH constituent stocks trade at forward PE ratios between 22x and 38x, compressed from mid-2024 peaks but still elevated relative to historical semiconductor cycles. Third, Microsoft, Meta, and Google have collectively announced $150 billion in 2025 capital expenditure plans, with roughly 60% earmarked for AI infrastructure. Aschenbrenner's short implies either demand destruction from model efficiency gains, supply saturation from TSMC's Arizona and Taiwan fab expansions, or margin compression as hyperscalers negotiate volume discounts on next-generation Blackwell chips.
The disclosure arrives as other AI-native researchers diverge on semiconductor exposure. Demis Hassabis maintains DeepMind's partnership with Google's TPU roadmap. Ilya Sutskever's Safe Superintelligence Inc. has raised $1 billion without taking public positions on chip supply chains. Aschenbrenner's bet stands apart because it contradicts his own published forecast that AGI-timeline acceleration would tighten compute constraints through at least 2027. The 13F suggests his private view shifted between June essay publication and Q4 portfolio construction, potentially based on proprietary data on training run costs, inference optimization, or hyperscaler procurement pipelines not visible in public filings.
Allocators should monitor three specific developments. NVIDIA's April earnings call will clarify whether Blackwell shipment cadence matches the $10 billion revenue guidance for fiscal Q2. Taiwan Semiconductor's March investor day will detail CoWoS packaging capacity additions, which determine how quickly NVIDIA can ship systems integrating HBM3e memory. And Microsoft's OpenAI partnership structure revision, expected by June, may signal whether the largest AI compute buyer sees path to lower per-token costs through architectural changes rather than chip volume. Each of these events will test whether Aschenbrenner's short thesis relies on near-term price action or structural overcapacity.
The 13F filing lists no equity longs, no credit positions, and no disclosed exposure to power infrastructure or data center REITs. Situational Awareness appears built as a concentrated short book, which means the fund likely runs low net exposure and depends on timing semiconductor sector repricing within a specific window. Aschenbrenner now trades against the companies supplying the compute he once argued would become the century's most constrained resource.