Today was less about new buys than an unusually frank self-audit, with one fresh piece of research stapled on.

A portfolio reckoning. In a long retrospective, @aleabitoreddit walked through his three core themes — neoclouds (AI cloud capacity), memory, and photonics — and argued his blended book is "overwhelmingly green," with most long ideas up triple digits year-to-date. He singled out his top performers in each bucket: NBIS (Nebius, his favored AI-cloud long), EWY (an iShares South Korea ETF he treats as a leveraged Samsung/SK Hynix memory bet via long-dated calls), and SIVE (Sivers, the tiny Swedish indium-phosphide laser maker that is his single highest-conviction "light-source chokepoint" for co-packaged optics).

He was equally open about the misses. A disputed analyst report claiming CPO (co-packaged optics) delays — which NVDA publicly refuted — hit several of his small Taiwanese photonics names (Foci, MSSCorp, Xintec), leaving them deep in the red, though he said Shunsin and Win Semi held up better. Korean supply-chain names (Auros, Foosung) gave him what he called lasting "PTSD," and he hinted he may avoid the volatile Korean small-caps entirely. He also conceded XLU, his utilities/AI-power-grid trade, suffered after the Iran war erased rate-cut odds, and that software names TTD and META went red. On SNAP, he reaffirmed his soured stance, citing aggressive stock-based-comp accounting. The bright spot among his "money printer" longs: RDDT (Reddit), finally green after a long drawdown.

On the winners' side he reiterated a broad roster: MRVL, ARM, INTC, MU and large-cap memory on one end; small-caps AXTI (indium-phosphide substrates), LITE, AAOI, RPI, IQE (epiwafers) directionally working; plus SIMO, HPS.A (transformers), TSEM (a photonics foundry), AEHR (burn-in test), SOI and ALRIB (Riber, MBE equipment). His caveat was about entry price — he flagged that someone buying AAOI near $220 rather than his cited $30 would be sitting on losses, and stressed judging ideas over months, not weeks.

The actual news: LPK. The day's most substantive item was his takeaways from meeting notes on LPK, his glass-core-substrate equipment play (the patterning tools for next-generation chip packaging are a chokepoint he's flagged repeatedly). He said the company is actively weighing a Nasdaq listing, is targeting roughly 70% market share, sees its addressable market running well above prior projections, and expects four to five customers placing orders this year as the industry readies high-volume ramps. He called it undervalued on these disclosures and confirmed he holds it.

A regret he won't repeat. He lamented floating a thesis on MLCC (multilayer ceramic capacitor) supply bottlenecks without buying — Taiyo Yuden, Murata and VSH (Vishay) have since surged 150%-plus in two months — and joked he should simply go long whenever he names a chokepoint.

Asides. He mused about how China keeps acquiring school-bus-sized ASML lithography machines despite export controls, and raised a structural worry about "content-farm" sites flooding low-quality AI-generated news that then pollutes large language models — a feedback loop he says causes basic technical errors, like conflating a single channel of a CW DFB laser array with a high-power single laser. He also took a swipe at a Bernstein call that turned bearish on Kioxia just before the stock rallied.

*Derived commentary, not advice.*