Serenity, the small-cap stock-picker whose reconstructed views this site tracks, surfaced briefly on Sunday to note he was feeling under the weather and taking a short break. Even so, he shared a batch of observations from the AI supply chain he watches obsessively — most of them reinforcing his core idea that the real money in the AI buildout sits in the upstream chokepoints, the scarce components that everything else depends on.

The laser land-grab. The headline item was channel chatter from ECOC (an optical-networking industry conference) suggesting Nvidia has locked up roughly half of Lumentum's new Greensboro capacity for ultra-high-power (UHP) lasers, with contract terms that could let it claim the rest. Nvidia (NVDA) is the AI accelerator leader whose roadmap Serenity treats as a demand telegraph; Lumentum (LITE) is the large-cap optical-components incumbent he holds at a roughly 5% weight as the "safe" anchor of his photonics sleeve. He read the move as classic allocation game theory: by hoovering up scarce lasers — much as it has with HBM memory — Nvidia makes it structurally harder for rivals to scale their own optical programs. His takeaway was constructive for Lumentum's visibility and, more importantly, argued it raises the strategic value of the smaller merchant laser suppliers he actually chases for upside.

Substrate bottlenecks. He linked news that Samsung Electro-Mechanics is expanding FC-BGA substrate capacity — notably funded by customers — to his packaging thesis. FC-BGA is a high-end chip-packaging substrate, and Serenity recalled that Marvell (MRVL) has cited large-body substrates in SEC filings as a core bottleneck of its custom-ASIC program, alongside sole-source dependence on TSMC. Marvell is the "does-everything" custom-silicon and optics play he holds as a mega-cap satellite; TSMC (TSM), the indispensable foundry he sizes at about 5% and calls the cleanest proxy for the whole AI trade. He noted Marvell has been striking its own supply agreements (referencing AT&S) to secure that capacity.

AMD and physical AI. He flagged AMD's reported $8.2B acquisition of World Labs, tying it to his long-held view that "physical AI" and world models would be a major shift — and ruing that he never played it via private markets. AMD (Advanced Micro Devices), which he does not own personally, functions in his framework mainly as another demand engine pulling through his upstream picks.

Odds and ends. He viewed Nvidia's roughly $150B buyback authorization as evidence of how much cash the company is generating, drawing an Apple (AAPL) comparison from its buyback era. And on OpenAI reportedly delaying a model release over safety concerns — alongside Nvidia launching an AI-safety platform — he shrugged it off, quipping that models getting "too good" is hardly a bearish signal.

*This is derived commentary for information only, not investment advice.*