Earnings season handed Serenity, the stock-picker tracked here for his bet on the obscure upstream suppliers feeding the AI optical buildout, a batch of results he read as broad validation of his thesis. His method is to treat hyperscaler spending as a signal of where demand lands, then buy the smallest "chokepoint" supplier — a hard-to-replace vendor with few substitutes — the giants quietly depend on.

AXT and the InP chokepoint. The centerpiece was AXT (AXTI), a maker of indium phosphide (InP) substrates — the compound-semiconductor wafers used to build high-speed lasers for optical links — and Serenity's highest-conviction holding. He flagged record quarterly revenue of $47.6M, with InP sales hitting a record $30.7M driven by data-center demand, revenue up 164% year over year. Management guided InP capacity toward $60M per quarter exiting 2026 and $130M exiting 2027, plans to double capacity in 2026 and again in 2027 to become the world's largest InP producer, and pointed at gross margins starting with a "five." He noted that demand keeps outpacing supply, that current strength comes from 800G/1.6T links with CPO (co-packaged optics, which places the optical engine next to the chip) extending the cycle past 2027, and that this doesn't yet include the price hikes he expects as substrates tighten. Notably, he pointed out this quarter didn't materially reflect agreements with peers Coherent (COHR) and Lumentum (LITE), leaving further upside. Coherent, a vertically integrated photonics supplier, is one of his steadier core holdings.

Hyperscaler capex confirms the demand side. Serenity read Amazon (AMZN) as strongly bullish for the AI-semi trade: it lifted 2026 capex to roughly $220B, said even that won't meet demand through 2027, and noted 2027 and much of 2028 capacity is already reserved. Higher memory costs baked into that number, he argued, are bullish for Micron (MU), his memory-supercycle anchor. He stressed that Alphabet (GOOGL), Microsoft (MSFT), Meta (META) and Amazon all cited a compute *shortage* — directly contradicting recent "excess compute" and "AI bubble" narratives he dismissed as noise.

XFAB confirms CPO and 800V exposure. Serenity said X-Fab (XFAB), a specialty foundry, explicitly confirmed on its call that it is progressing in co-packaged optics and expects the industry shift to 800V data-center power to lift demand for its silicon-carbide and gallium-nitride (SiC/GaN) chips — refuting bears who claimed the foundry had zero exposure to those trends. He cited three silicon-photonics projects and new SiC design wins, though photonics volume production isn't expected until 2028, slightly later than he'd hoped. Power-chip names Navitas (NVTS) and Power Integrations (POWI) surfaced only as XFAB-linked 800V beneficiaries, not as personal positions.

Sivers gets a nod. He revised up his read on POET Technologies (POET), an optical-packaging firm, noting its language shifted from "readiness" to "production" with lasers from Sivers (SIVE) — his largest and most-loved position, a tiny Swedish InP laser maker he views as the true upstream chokepoint. He expects volume orders flowing from Jabil (JBL) through POET to hit in the first half of 2027, consistent with Sivers' recent oversubscribed fundraise. As always, this is derived commentary, not advice.