Serenity, the small-cap stock-picker whose framework hinges on buying the obscure upstream suppliers that feed the AI buildout, spent Friday building his case for his single largest conviction position and reinforcing his memory thesis.

The Sivers capacity argument. The bulk of his day centered on Sivers Semiconductors (SIVE), the tiny Swedish maker of indium-phosphide (InP) continuous-wave lasers — the light source that silicon photonics and co-packaged optics (CPO) depend on. His read: the industry is scrambling to lock up any qualified CW laser capacity, and the traditional merchant suppliers have quietly pulled their output in-house. He pointed to Coherent (COHR), a vertically integrated photonics leader spanning substrates, lasers and transceivers, saying it has no near-term ability to sell InP lasers externally because internal demand consumes it all. Applied Optoelectronics (AAOI), the Made-in-America transceiver maker he also owns, echoed the same dynamic, while Lumentum (LITE), his large-cap incumbent benchmark, has reportedly been buying lasers on the open market to relieve its own bottleneck. His conclusion: with merchant capacity vanishing, big transceiver names must source elsewhere — leaving room for Sivers to step in with fresh foundry capacity via Win Semi and a second, already-qualified partner.

He seized on Sivers' latest earnings signals: six new pluggable/module engagements and a roughly doubled opportunity pipeline. He argued the pipeline jump dwarfs what the disclosed Jabil (JBL) and GlobalFoundries (GFS) wins alone would imply — Jabil being the contract manufacturer that chose Sivers' lasers for its 1.6T pluggables, and GFS a US photonics foundry. On why customers aren't named or locked into long-term agreements yet, he leaned on the Marvell (MRVL)–POET Technologies (POET) precedent: vendors often can't be disclosed, and qualification cycles can't simply be skipped. POET, the optical-packaging firm he views as merely packaging lasers sourced from upstream players like Sivers, again served as a foil rather than a pick. He listed a wide roster of potential Sivers customers and concluded the shortage conditions materially raise his odds that engagements convert to revenue.

Memory supercycle, reaffirmed. Serenity also refreshed his structural memory-shortage thesis. He cited SK Hynix (SKHY) guiding to a memory shortage lasting through 2030, Sandisk (SNDK) — his core NAND long — pointing to structural demand and ~80% gross margins, and NVIDIA (NVDA) raising its supply commitments from roughly $119B to $279B, largely on memory purchasing. NVDA is his demand telegraph, not a holding; he front-runs its suppliers instead. His takeaway: forward valuations across the memory sector have room to expand further.

Lighter posts rounded out the day — a wish for hedge-fund-sized firepower to buy the dip on "stupid commentary," and a musing that a hypothetical $2T Anthropic IPO could buy up dozens of consumer brands with a fraction of its value. This is derived commentary, not advice.