A hedge fund run by a former OpenAI employee disclosed options positions worth roughly $47 million notional exposure across Nvidia and AI semiconductor names in its latest 13F filing, filed with the SEC on May 15th. The fund, which launched in late 2023 with $120 million in initial capital from two family offices and a university endowment, dedicated 39% of its reportable book to derivatives tied to chip supply-chain consolidation.

The filing shows call options on Nvidia representing $32 million in notional value, with strikes ranging from $900 to $1,150 and expirations clustered in September and December 2025. Smaller but notable positions include Taiwan Semiconductor Manufacturing, Broadcom, and ASML, each with options books between $3 million and $6 million. The fund holds no puts on any semiconductor name. The manager, who spent three years at OpenAI working on infrastructure procurement before departing in mid-2023, filed the fund's ADV in November with $180 million in assets under management by year-end.

The timing matters. Nvidia's Blackwell architecture, delayed twice in 2024, is now scheduled to ship in volume during Q3 2025. TSMC's Arizona fab—producing the same 4nm process used in Blackwell's compute tiles—reached 70% yield parity with Taiwan in April, ahead of its July ramp target. Broadcom, meanwhile, reported $12 billion in custom AI chip backlog during its March earnings call, a 340% increase year-over-year. The options expirations align with Blackwell's expected deployment at hyperscalers and the first full quarter of Arizona fab output.

This is not a passive bet on AI hype. The fund's structure suggests conviction on a narrower thesis: that compute bottlenecks force hyperscalers to consolidate spend among the three vendors capable of delivering at scale—Nvidia for GPUs, Broadcom for custom ASICs, TSMC for leading-edge packaging. The absence of diversification into software or application-layer AI names is deliberate. The fund's ADV lists two full-time analysts, both with backgrounds in semiconductor supply-chain logistics, not machine learning.

Operators and allocators should watch TSMC's June earnings call for Arizona yield guidance, Nvidia's July shipment cadence during its fiscal Q2 report, and Broadcom's June analyst day for custom silicon margin disclosure. Family offices eyeing similar derivative structures need clarity on whether those September Nvidia calls roll forward or close ahead of Blackwell's actual data-center deployments, likely visible in October hyperscaler capex filings. The fund's next 13F, due mid-August, will show whether it added to positions during May's semiconductor selloff or trimmed ahead of macro uncertainty.

The former OpenAI employee has not spoken publicly about the fund's strategy. The ADV filing lists no marketing materials and no investor calls. The $47 million in notional exposure is not the number. The number is three—the vendors left standing when compute moves from experiment to infrastructure.