← All commentary
2026-10-19

Serenity reads a Lumentum CEO interview as proof the photonics demand wall is still standing

On a day when broad market weakness dragged AI names down together, Serenity — the stock-picker known for back-mapping Nvidia's roadmap to the obscure upstream suppliers that feed it — argued investors were throwing out good names with bad ones.

Photonics: demand versus financing risk. His first focus was Lumentum (LITE), the large-cap optical-components incumbent he holds as the "safe" anchor of his photonics sleeve, tied to Google TPU optics. Parsing a CEO interview, Serenity highlighted management's claim that hyperscaler capex is still rising, that its purchase agreements are largely non-cancellable, and that it now has capacity committed into 2029 with 2030 under negotiation. His takeaway: demand and financing risk are not evenly spread across the AI supply chain, yet the market is selling everything in unison. He reiterated his structural point that rising optical content per XPU plus adoption of co-packaged and near-packaged optics (CPO/NPO — putting the optical engine next to the switch die to save power) should lift photonics spending even if aggregate capex dips. He framed Lumentum as less dependent on pure datacenter buildout than networking rival Broadcom (AVGO).

SIVE as the upstream wildcard. Serenity then returned to his highest-conviction small-cap, Sweden's Sivers Semiconductor (SIVE), which makes the continuous-wave lasers that serve as light sources for silicon photonics and CPO — a classic "chokepoint," a scarce upstream part the whole buildout leans on. He reviewed its disclosed 100MW CW laser capacity target for Q4 2027, a second foundry partner alongside Win Semiconductors, and expected revenue from contract manufacturer Jabil (JBL) in 2027. He cautioned that, unlike Applied Optoelectronics (AAOI) — the Made-in-America transceiver maker he follows closely — Sivers gives too little product-level detail to model precisely, so the bet rests on whether qualifications convert, the same risk he watched play out at burn-in test firm Aehr (AEHR). He noted Sivers appears as a reference laser in GlobalFoundries' (GFS) CPO platform, and that downstream packagers like POET and LiDAR maker Aeva (AEVA) are smaller nodes in the same chain. He stressed Sivers uses external fabs and is already institutionally financed, so it doesn't share the funding worries hitting other corners of AI.

Neoclouds: a cancellation he read as bullish. Finally, Serenity pushed back on the consensus reaction to Microsoft (MSFT) walking away from a deal with NScale, reportedly over UK grid constraints. His logic: if Microsoft still urgently needs capacity it can't get from one provider, it will buy elsewhere — making rival "neocloud" operators with energized capacity more valuable, not less. He named his core long Nebius (NBIS), the best-financed neocloud in his view thanks to Nvidia backing, while noting the episode also warns what happens when a provider can't deliver contracted demand — a dig at Iris Energy (IREN), his structural short over its large dilutive equity offering.

This is derived commentary, not advice or his own words.

Generated by the claude-opus-4-8 pipeline. Derived content; not investment advice.