Serenity, the small-cap stock-picker known for front-running the AI photonics buildout by buying obscure upstream suppliers, spent Monday taking stock. After a recent market pullback, he counted which of his calls were still up more than 100% year-to-date — and landed on roughly 19 names spread across his usual themes.

The list read like a tour of his book. In memory, his structural "supercycle" thesis of AI-driven DRAM and NAND shortages, he flagged MU (Micron, his US-made memory anchor), SNDK (Sandisk, his NAND pricing-power play via Kioxia), and SIMO (Silicon Motion, a backdoor controller name), alongside EWY, the South Korea ETF he treats as a leveraged Samsung/SK Hynix volatility trade. In photonics — his dominant theme of light sources and substrates feeding co-packaged optics (CPO), where optical engines sit beside the switch or GPU die — he cited SIVE (Sivers, the tiny Swedish indium-phosphide laser maker he calls his top pick), AXTI (AXT, InP substrates), SOI (Soitec, silicon-on-insulator substrates), IQE (the UK epiwafer foundry), plus merchant names LITE (Lumentum) and AAOI (Applied Optoelectronics). MRVL (Marvell) appeared for its custom AI ASICs, INTC (Intel) for foundry, NBIS (Nebius) for his neocloud long, AEHR (burn-in test) for packaging, ARM for his inference-era CPU thesis, and RPI for AI orchestration. GDRZF (Gold Reserve) rounded it out on a Venezuela arbitration angle — a name he lists as a winner but has never built a standalone thesis around.

Notably, he was candid that several core holdings had slipped out of the 100%-plus club, including foundry pick TSEM (Tower Semiconductor, his "TSM of photonics"), glass-core substrate equipment maker LPK (LPKF, his near-monopoly "ASML of glass substrates"), and power name HPS.A. He also admitted disappointment in laggards like XFAB (X-Fab) and Shunsin. The framing matters: many of these are early-innings, catalyst-driven bets — LPKF's glass-substrate ramp isn't expected until 2026-2027 — so short-term underperformance doesn't dent his thesis, but he's not hiding it either.

He noted, with some amusement, that Chinese-language followers have coined a name for his style — a "Perilla Leaf Theory," a nod to the herb that quietly turns up everywhere in a dish, much as his chokepoint suppliers sit hidden upstream of the big AI names.

The forward-looking news was two new H2 ideas. He introduced CCXI as a play on humanoids and the shift toward "physical AI" — downstream robotics demand rather than his usual data-center plumbing. And he named ESMT (a Taiwan-listed memory maker) for legacy DDR2/DDR3 chips, a niche corner of the memory market where older-node scarcity can create its own pricing squeeze.

Serenity offered his customary caveat that he can't promise the new picks repeat the run, and said he'll keep publishing his reasoning openly for others to challenge. Today was less a trade than a scorecard — and a preview of where he's hunting next.

*Derived commentary, not investment advice.*