The Bucket
A single name — Sandisk (SNDK) — at ~4.5 % of the Q4 2025 13F. The bucket is small but thematically distinct: it is the fund’s only direct memory/storage exposure, and it expresses a specific bet on NAND flash as opposed to HBM/DRAM.
Thesis
Three structural shifts make NAND interesting in the AI cycle:
- Inference-time KV-cache persistence. Long-context inference creates pressure to swap KV state out of HBM and onto fast NAND tiers (NVMe-attached or CXL-attached). Per-token economics favor cheaper-but-fast storage where latency permits.
- Data-set scale. Frontier labs increasingly hoard multi-petabyte training datasets that need fast NAND for repeated epoch reads.
- Supply discipline. NAND is a commodity oligopoly (Samsung, Kioxia/Sandisk, SK Hynix, Micron, YMTC). After the 2023–2024 down-cycle, surviving suppliers cut capex sharply. Prices recovered in 2025, and the AI-driven step in demand could keep the cycle elevated through 2027.
Sandisk specifically is a pure-play NAND name following its February 2025 spinoff from Western Digital (WDC was held by the fund in Q3 2025 and exited in Q4 in favor of the cleaner SNDK expression). The fund explicitly owns the storage company and not the HDD company that WDC kept.
What’s Not in the Bucket
- No Micron (MU). The fund actually held MU puts in Q3 2025 (closed by Q4) — explicitly negative on DRAM/HBM near term. MU’s HBM3e share, while important, is consensus-loved.
- No SK Hynix or Samsung. Non-US listings outside the 13F universe.
- No Pure Storage / NetApp / Hammerspace / WekaIO. The fund prefers raw-component exposure to system-vendor exposure.
Constituents
| Company | Ticker | Q4 2025 value | % of 13F |
|---|---|---|---|
| Sandisk | SNDK | $250.2 M | 4.53% |
Position History
WDC was added in Q3 2025 ahead of the spin completing. Q4 2025 rotated the position into the spun-out SNDK at increased size, while exiting the residual WDC. Net effect: cleaner pure-play exposure to NAND with no HDD legacy.