A hedge fund managed by a former OpenAI executive disclosed a $13.7 billion portfolio in its latest 13F filing, structured around short exposure to semiconductor manufacturers through put options on SanDisk and Nvidia, while establishing long positions across energy sector equities. The filing, which surfaced this week without naming the fund or its principal, marks one of the largest directional bets against chip stocks filed by a technology-adjacent fund in the current cycle.
The portfolio's architecture is exact: put option concentrations on semiconductor names that have appreciated 180% to 240% since January 2023, combined with outright equity stakes in energy companies that have lagged the S&P 500 by 14 percentage points over the same period. The SanDisk position is notable—puts on a Western Digital subsidiary trading at $87 per share, up 61% year-to-date, suggest a thesis targeting storage economics rather than compute. Nvidia puts, sized to hedge a notional exposure in the hundreds of millions, indicate conviction that datacenter capex will decelerate faster than current analyst models project. The energy longs span upstream producers and midstream infrastructure, not renewables.
This matters because the position represents a second-order view on AI infrastructure costs. The trader left OpenAI in late 2023, after the company's annualized revenue run rate crossed $3.4 billion and its compute spending was rising at 40% quarter-over-quarter. The fund's construction implies a belief that AI model training costs will compress—either through algorithmic efficiency gains that reduce chip demand, or through energy constraints that limit datacenter expansion faster than semiconductor supply can adjust. Energy longs hedge both scenarios: if power becomes the binding constraint for AI scaling, energy equities appreciate; if chip demand craters, energy correlates less with technology drawdowns. The SanDisk exposure suggests the thesis extends to storage arbitrage, where NAND pricing has remained elevated despite declining utilization in hyperscale environments.
Allocators should monitor three signals. First, whether Nvidia's datacenter revenue growth decelerates below 25% quarter-over-quarter in its January 2025 earnings report, due late February—a break in consensus that would validate the put thesis. Second, power purchase agreements from hyperscalers, particularly Microsoft and Meta, which have committed to 15 gigawatts of new datacenter capacity through 2027; any deferrals or renegotiations would pressure chip demand forecasts. Third, whether other AI-adjacent funds file similar 13F positions in the current cycle—the deadline for Q4 2024 filings is February 14, 2025. One former OpenAI principal positioning this way is data; three or four would be a pattern.
The fund has not disclosed leverage ratios, but put option notional values imply balance sheet capacity in the $20 billion to $30 billion range if the positions are hedged with cash-secured strategies. The energy longs are not paired with corresponding shorts, which means the fund is running net long equity exposure while expressing a directional short view on one sector. That structure works if energy underperformance reverses, or if semiconductor weakness arrives faster than broad market repricing.