Serenity, the small-cap-focused investor whose reconstructed views this site tracks, spent Monday doing two things: rebutting the growing "AI bubble" narrative, and pointing to a little-known memory company's earnings as evidence that the demand underneath the trade is real, not hype.

The anti-doom argument

Serenity's central framework is "buy what's upstream of Nvidia" — back-mapping the AI buildout to the small, hard-to-replace suppliers of lasers, substrates, memory and optics that the hyperscalers quietly depend on. Today he argued that the loudest voices on social media have simply discovered that bearish posting earns more attention, and that the underlying contracts tell a different story.

He cited a cluster of his own themes as counter-evidence. SpaceX (SPCX) and Elon Musk building aggressively in the Neocloud (AI cloud capacity) space, he argued, signals they see real returns — SpaceX's anticipated mega-IPO is a theme he watches for its market-wide liquidity ripples. He noted Samsung signing five-to-ten-year memory supply agreements with Chinese tech giants, which he reads as insiders betting there will be no glut. And he pointed to Lumentum (LITE) — his large-cap benchmark in the AI photonics sleeve, a supplier of optical components for Google's TPUs — being sold out into 2029 and announcing new ultra-high-power laser contracts. His pointed jab: the same commentators who were long memory maker Sandisk (SNDK), Neocloud favorite Nebius (NBIS) and vertically integrated photonics leader Coherent (COHR) just two months ago are now the ones calling a bubble. He contends the real froth sits in triple-digit-forward-P/E cybersecurity and biotech names, not in memory companies trading near 3x earnings with hyperscaler long-term agreements locked in.

A memory name he thinks the market is mispricing

The sharper note came on ESMT — Elite Semiconductor Microelectronics, a roughly $2.3B Taiwanese maker of niche DRAM (including legacy DDR2/DDR3) and specialty memory. The company isn't a core holding he's disclosed, but it fits his NAND/DRAM supercycle thesis: the view that AI-driven demand has pushed memory into a multi-year supply crunch with repeated price hikes. He walked through quarterly numbers showing roughly $679M revenue at a ~77% gross margin and nearly 59% operating margin, and flagged a large net-cash and inventory cushion — enough, on his math, that the firm would accumulate assets exceeding its market cap within about a year. He also relayed management commentary (via Economic Daily) that no meaningful jump in DDR2/DDR3 supply is expected in 2027, implying prices hold, with further hikes anticipated in Q4.

His takeaway was that a company trading at roughly 1.6x annualized operating earnings selling off is the market "trolling" — exactly the kind of deep-value, catalyst-backed dislocation his approach hunts for.

*This is derived commentary for information only, not investment advice, and does not represent Serenity's own words.*