Serenity, the small-cap stock-picker who front-runs the AI buildout by buying the obscure upstream suppliers — the lasers, substrates and test tools — that the big chip names quietly depend on, had two distinct threads today: a deep-value legacy-memory idea and a fresh reading of the optical supply chain straight from TSMC.

A legacy-memory anomaly

His memory theme rests on the idea that AI is dragging the whole DRAM/NAND market into a multi-year supply squeeze, lifting prices on even old, "legacy" chips. Today he turned to Taiwan-listed ESMT (3006), a maker of legacy DRAM and NOR/SLC NAND flash. Serenity's argument is arithmetical: stripping out net cash and inventory, he reckons the company trades at roughly 1.9x its July earnings run-rate — an outlier when, by his account, most memory-linked names are being bid up to around 20x forward 2028 estimates. He noted analyst expectations for average selling prices to jump more than 100% quarter-on-quarter and sees legacy DRAM/NAND shortages tightening through 2027, while conceding the key uncertainty is how long the upcycle lasts and whether wafer-cost hikes get passed through. He framed it as an observation rather than a call, leaving the verdict to the market.

TSMC hands him a bottleneck map

The second post keyed off comments from a TSMC (TSM) advanced-packaging executive. TSMC is the foundry every AI chip ultimately runs through, and Serenity holds it as a safe anchor position — so when its packaging boss says the real constraints on large-scale co-packaged optics (CPO, the technique of placing the optical engine right next to the switch or GPU die) are not the optical engine itself but lasers, fiber, connectors and test, that reads to him as a roadmap of where demand lands.

He sorted the beneficiaries into buckets. On lasers — the light sources for high-speed optical links — he listed incumbents Lumentum (LITE) and Coherent (COHR), both of which he owns as lower-beta core photonics holdings, alongside his highest-conviction bet, Sweden's Sivers (SIVE), a tiny maker of indium-phosphide lasers he treats as the overlooked chokepoint of the whole CPO wave. He also cited Taiwan's LandMark (3081), the capacity-constrained pure-play he uses as a valuation yardstick.

On fiber and connectors, he pointed to Corning (GLW), the large incumbent in glass and optical fiber he respects but tends to skip in favor of smaller upstream names. On the test-and-measurement layer — which he views as its own scaling bottleneck as optical parts need validation — he named FormFactor (FORM), Keysight (KEYS) and Taiwan's MPI (6223).

He stressed nothing here is dramatically new versus the first half, and that the specific companies were his own additions, not endorsed by TSMC. His point was simply that periodically restating who benefits keeps the thesis honest — and that a demand imbalance in fiber-array units flagged for 2027 keeps the pressure building upstream.

*Derived commentary, not investment advice.*