A quiet session for Serenity, the stock-picker tracked here for his bet that the AI boom's real money sits in the obscure upstream suppliers that feed the big chipmakers. His single comment today pushed back on a round of media coverage of OpenAI's finances and argued the chatter has nothing to do with his core theses.

The news he was reacting to. Headlines circulated comparing OpenAI's annualized revenue — which Serenity pegged at roughly $50 billion — with higher figures he attributed to flawed accounting comparisons against rival Anthropic. The implication some readers drew was that softer or disputed AI-model revenue might undercut the hardware buildout. Serenity's response, paraphrased: it doesn't, and the logic doesn't connect.

Why it doesn't touch his thesis. His framework is supply-chain back-mapping — he treats the demand roadmap of the hyperscalers and Nvidia as a telegraph for where orders will land, then buys the smallest "chokepoint" suppliers (a critical upstream vendor with few substitutes) that the whole buildout depends on. In his view, a debate over one AI lab's top line says little about the multi-year physical shortages he's playing, which are driven by capacity constraints in lasers, optics and memory rather than by any single customer's revenue.

He pointed to two core holdings as the examples.

- AAOI (Applied Optoelectronics) makes optical transceivers — the components that turn electrical signals into light to move data between servers in a data center. Serenity frames it as the only pure "Made-in-America" vertically integrated optical play, doing laser fabrication, design and assembly in-house in Texas, and argues hyperscalers are buying all the 800G/1.6T capacity it can produce as part of a three-to-five-year laser shortage. That supply crunch, he says, is a physical bottleneck no revenue headline resolves.

- MU (Micron Technology) is the US memory maker and the centerpiece of his "memory supercycle" thesis — the idea that AI demand has created a structural shortage in DRAM, high-bandwidth memory (the stacked DRAM sitting beside GPUs) and NAND flash, lifting prices and margins. He has previously disclosed a roughly 10% portfolio weight to it. Again, his point today was that a shortage measured in years of capacity doesn't end because of an accounting dispute over a customer's sales figure.

The takeaway. Nothing in Serenity's positioning changed today; the note was a defensive clarification rather than a new move. His message was consistent with his longstanding stance: the investable constraint in AI is physical supply of lasers, optics and memory, and that bottleneck is decoupled from the week's noise about model-maker economics. He named no new tickers and signaled no trades.

This is derived commentary for context, not investment advice, and reflects a paraphrase of his public posting rather than his exact words.