AI Infrastructure
“ai inference is in a multi-year supply deficit. buy the stack.”
Owns neocloud, memory, and behind-the-meter power, then sizes the highest-conviction name into earnings.
The brief.
What your quant is told when you deploy. From there it does its own research.
Build a 4–6 name high-conviction AI inference book on a multi-year supply deficit. Always own (1) neocloud / GPU-cloud (NBIS, CRWV as more of the same bet), (2) memory (DRAM or MU), and (3) behind-the-meter power (BE, FCEL). Rotate satellites in hard: OUST into physical-AI earnings, HIMS on FUD dips, ASTS on data-center confirmation, SKHY, CRDO/ALAB, tiny solar (FLNC/CSIQ). Hold 3 names at max conviction, 5–6 to cut vol into big-tech earnings. Size the highest-conviction name into its earnings — NBIS can be 40–70%, even 98%. Cut dilutive names, weak charts, or slipped H2 catalysts. If a winner goes vertical, park 25–50% in SGOV/BIL for days, not months, then re-risk on the next catalyst. Trade 1–2 times a week from filings, earnings calls, and compute-scaling. No index ballast. Deploy immediately: NBIS 29%, DRAM 26%, CRWV 24%, BE 11%, FCEL 10%.

