A hedge fund managed by a former OpenAI executive disclosed a $13.7 billion portfolio weighted toward semiconductor put options and energy infrastructure positions in its latest 13F filing. The fund, whose principal left OpenAI within the past eighteen months, structured the portfolio with significant downside protection on leading chip manufacturers while concentrating long exposure in natural gas pipelines and grid-scale power assets. The filing marks one of the larger AI-adjacent capital deployments to take an explicitly defensive posture on the semiconductor supply chain.
The put positions target companies across the AI chip stack, with notable concentration on firms supplying advanced packaging, memory controllers, and datacenter networking silicon. The energy positions favor regulated utilities with multi-GW datacenter contracts and independent power producers tied to natural gas availability in Texas and the Mid-Atlantic. Portfolio construction suggests the manager expects either margin compression in AI-specific semiconductor segments or physical bottlenecks in power delivery to hyperscale facilities. Both outcomes would validate a thesis that AI infrastructure returns accrue to energy providers rather than chip designers.
This positioning diverges from consensus allocator behavior in two dimensions. First, institutional money has treated semiconductor exposure as a direct AI beneficiary trade since mid-2023, not a volatility hedge. The put structures imply the manager sees either valuation risk or competitive compression as new entrants—particularly vertically integrated cloud providers designing custom silicon—erode third-party chip margins. Second, the energy concentration bypasses datacenter REITs and colocation operators in favor of upstream power suppliers, suggesting the thesis centers on electricity scarcity rather than rack space. That view aligns with forward power purchase agreements now extending 18-24 months out in major datacenter markets, a timeline that makes regulated utilities with locked contracts more attractive than speculative development plays.
The filing's structure also reveals execution sophistication. The put positions appear calibrated to hedge long equity exposure elsewhere in the portfolio rather than express outright bearish conviction, with strike prices suggesting 15-25% downside protection on specific names. The energy positions favor assets with FERC-regulated returns or take-or-pay contracts, minimizing exposure to spot commodity volatility. This combination—hedged semiconductor exposure plus contracted energy cash flows—indicates the manager is positioning for a scenario where AI model training costs rise due to infrastructure constraints, not technology failures. If datacenter power costs double while chip performance gains slow, the portfolio benefits from both legs.
Allocators should monitor three forward signals. First, whether other AI-native operators begin filing similar defensive semiconductor positions in Q1 2025 disclosures, due by mid-February. Second, power purchase agreement pricing in Virginia and Texas datacenter corridors, where forward curves have already steepened 30-40 basis points since October. Third, custom silicon announcements from hyperscalers, particularly if Microsoft or Meta accelerate in-house chip timelines beyond currently disclosed 2026-2027 production starts.
The manager's OpenAI tenure ended before the organization's latest compute scaling announcements, but after the period when internal infrastructure cost models would have been visible. That timing matters. The portfolio reads less as market pessimism than as operational realism from someone who has seen the actual electricity bills.