Serenity spent Friday taking unusually candid stock of his own portfolio, walking through what has worked, what has hurt, and where he thinks he simply missed.

A rare honest scorecard. He reiterated that his book rests on three pillars — neoclouds/AI power, the memory supercycle, and AI photonics — and argued he landed on the standout in each: Nebius (NBIS), his favored AI cloud and "next hyperscaler" sum-of-parts long; EWY, the South Korea ETF he uses as a leveraged Samsung/SK Hynix memory proxy via long-dated calls; and Sivers (SIVE), the tiny Swedish maker of indium-phosphide CW lasers — the light source he sees as the chokepoint feeding co-packaged optics (CPO), where the optical engine sits next to the switch or GPU die. He framed his blended return as overwhelmingly green, with most ideas up triple digits this year, while stressing that not every name is green on short timeframes and that entry price matters enormously.

Where it hurt. He flagged real losses on a cluster of Taiwanese CPO suppliers — Foci, Microcosm and Xintec among them — after a since-refuted analyst report on CPO delays that Nvidia (NVDA) pushed back on. Shunsin and Win Semi held up better. Korea remains his sore spot: he cited pain in names like Auros and Foosung and said the volatility may keep him away, though he still expects the supply chain to recover on SK Hynix and Samsung qualifications. He also conceded XLU, his "AI power bottleneck" utilities trade, didn't pay off after the Iran conflict erased rate-cut hopes, and that a few software bets (TTD, META, SNAP) went red, with Reddit (RDDT) the bright spot.

The names that worked. On the large-cap side he pointed to Marvell (MRVL), the custom-ASIC and optical play he treats as the hub linking to SIVE; Arm (ARM), his inference-era CPU thesis; Micron (MU), the core memory long; plus Lumentum (LITE) and Intel (INTC). Among smaller caps he cited AXT (AXTI) — the InP substrate chokepoint that remains his most legendary call — alongside Applied Optoelectronics (AAOI), IQE (epiwafers), RPI and others, plus a grab bag including controller-maker Silicon Motion (SIMO), Tower Semiconductor (TSEM), burn-in test house AEHR, transformer maker Hammond Power (HPS.A) and SOI.

The fresh catalyst: LPK. The day's most actionable item came from new meeting notes on LPK, a glass-core substrate patterning-equipment maker — glass cores being a next-gen chip-packaging substrate whose tooling is a bottleneck. Serenity, who owns it, said he and the market both overlooked the takeaways: a Nasdaq listing is under active discussion, the company is targeting an aggressive 70% market share, its addressable market looks far larger than prior estimates, and he expects four to five customers placing orders this year. He called it clearly undervalued.

Side observations. He mused on how China keeps acquiring ASML (ASML) machines despite export controls, warned about AI "content farm" spam polluting model training and retrieval — including garbled technical claims about laser arrays — and needled Bernstein for a bearish Kioxia call that preceded a sharp rally. He also lamented not buying his own MLCC-bottleneck idea, noting Vishay (VSH) and peers have since roughly doubled or more.

*Derived commentary, not investment advice.*