The Bucket
A single name — CoreWeave — but it accounts for ~22 % of the Q4 2025 13F across common stock and call options, making it the largest single-issuer exposure in the fund. The bucket exists to capture the non-hyperscaler path that AI compute spend can take: rather than buying NVDA equity directly or buying Microsoft/Google/Amazon (whose AI revenue is diluted by their non-AI businesses), the fund owns the merchant GPU cloud that exists almost solely to rent NVIDIA capacity to model labs.
Thesis
CoreWeave’s revenue is roughly a clean derivative of three things:
- NVIDIA shipment availability — every H100 / H200 / GB200 / GB300 they can take, they take.
- Model-lab GPU demand — multi-year reserved-capacity contracts with Microsoft (which hosts OpenAI) and a fast-growing roster of frontier labs.
- Power-secured data-center capacity — the binding constraint on growth. CoreWeave is increasingly a co-investor in data-center sites, including via deals with Core Scientific and Applied Digital that the fund also owns directly.
This is the purest “Situational Awareness” expression in the book: if Aschenbrenner’s essay is right that compute scaling continues, CoreWeave’s revenue compounds; if it stalls, CoreWeave is left with overpriced inventory and stranded leverage. The fund owning both the common (long delta) and a much larger call-option position (convex, levered upside) is consistent with high conviction in the upside scenario.
Constituents
| Company | Ticker | Q4 2025 line(s) | Value | % of 13F |
|---|---|---|---|---|
| CoreWeave | CRWV | Common + Calls | $1,211.2 M | 21.96% |
Position History
CoreWeave was added in Q1 2025 (the quarter after its IPO) at small size. Through Q2 2025 the position was held flat, with puts briefly added in Q3 2025 as a hedge during a sharp post-IPO rally. By Q4 2025 the puts were closed and the call position was added at $774 M notional — the fund’s largest single options bet. The common position was also increased to ~6.1 M shares.
Why It’s Its Own Bucket
CoreWeave doesn’t fit the “BTC miner pivots to HPC” thesis (it was never a miner — it was an Ethereum mining operation that pivoted to GPU cloud in 2019 before the AI wave), it isn’t a hyperscaler, and it isn’t a chip designer. It is the only US-listed pure-play merchant GPU cloud at meaningful scale. The fund treats it as a category of one.