Ex-OpenAI Researcher Disclosed $13.7B Hedge Fund With Chip-Maker Puts Untested by Volatility

A hedge fund managed by a researcher who previously worked at OpenAI filed a 13F disclosure showing a $13.7 billion portfolio that includes substantial put options against leading AI chip manufacturers. The fund also holds concentrated positions in energy sector equities and SanDisk. The filing marks the first public view of the strategy, but the semiconductor short thesis has not yet encountered the supply glut or demand correction that would demonstrate its validity.

The 13F shows put positions against chip makers at a moment when AI infrastructure spending remains elevated and hyperscaler capital expenditure forecasts for 2025 continue to rise. The portfolio structure suggests a conviction bet that semiconductor demand tied to generative AI workloads will contract or that gross margins will compress faster than consensus expects. The energy holdings and SanDisk position indicate sector rotation rather than broad market pessimism. The former OpenAI affiliation raises questions about information asymmetry—whether the thesis relies on insider perspective about AI compute efficiency gains that could reduce chip consumption per unit of model performance.

The timing matters because semiconductor equities have not yet repriced for margin compression or order pushouts. Nvidia closed the most recent quarter with datacenter revenue up 17% sequentially, and TSMC raised 2025 capex guidance by $7 billion in January. If the fund's thesis depends on a near-term inflection in hyperscaler orders or a shift toward inference-optimized chips with lower gross-dollar consumption, the market has not yet provided confirmation. The structure of the puts—strike prices, expiration dates, notional exposure—remains undisclosed, so tail-risk hedging cannot be ruled out. A $13.7 billion portfolio with puts as a minor percentage of total notional would tell a different story than a portfolio where puts represent the primary conviction.

Allocators should note that the thesis carries embedded timing risk. Semiconductor cycles historically turn on inventory digestion and capex curtailment, not on incremental efficiency in end-use applications. If hyperscalers continue to build out AI infrastructure ahead of revenue monetization—treating capex as a strategic moat rather than a margin decision—then order volumes could remain elevated even as performance per watt improves. The SanDisk holding suggests the manager sees value in legacy storage or expects a pivot toward edge inference requiring higher storage density. Energy positions could hedge against grid-capacity constraints that bottleneck datacenter expansion, but that thesis would support chip demand rather than undermine it.

Watch for two developments in the next 90 to 180 days. First, Nvidia's April earnings call will clarify whether Blackwell production is supply-constrained or whether customer order patterns are shifting. Second, any 13F amendment or Form 4 filing from the fund will show whether the put positions were reduced, expanded, or rolled forward after the initial disclosure. If the positions were closed shortly after the filing date, the strategy was likely a short-term volatility trade rather than a structural bear thesis.

The filing introduces a named contrarian position into a market structure that has priced semiconductor equities for sustained AI infrastructure buildout. The thesis will be tested only when hyperscaler capex growth decelerates or when gross margins compress faster than buy-side models anticipate—and neither has occurred yet.