Zhongji Innolight’s push toward an 8 billion dollar Hong Kong listing, potentially the city’s largest since Alibaba’s 2019 debut, reads on the surface like a new Chinese challenger stepping onto Marvell’s turf. The supply chain says otherwise. Innolight is the world’s largest optical transceiver assembler and a primary supplier to Nvidia and Google, but it does not design the digital signal processor silicon that sits inside its own modules. That layer is a duopoly held by Marvell and Broadcom, and there is currently no mature Chinese domestic alternative to either of them. Innolight is a customer in this chain, not a competitor to it.
The chip layer is where the margin actually sits
Seven of the world’s top ten optical module suppliers by shipment volume are Chinese, and Innolight alone is estimated to hold roughly half of Nvidia’s optical module wallet and as much as 80 percent of Google’s orders for modules above 800G in 2026. Being first in shipment share has not translated into first in profit share, because the DSP chips and EML lasers inside those modules still belong to Marvell, Broadcom, Lumentum, and Coherent. Broadcom’s Sian2 DSP is described as the standard inside Nvidia’s supply chain and is the part most commonly found inside Innolight and its closest rival, Eoptolink. Marvell’s exposure to Innolight runs through a different channel, most visibly a coherent module partnership built around Marvell’s Deneb DSP for telecom and 5G backhaul, while Marvell’s PAM4 families, Spica for 800G and Ara for 1.6T, skew more toward Microsoft and Google’s own infrastructure builds. Innolight’s AI-driven growth, in other words, has leaned more directly on Broadcom’s silicon than on Marvell’s.
What the capital raise actually does
A better-capitalized Innolight expanding 400G, 800G, and 1.6T capacity is a volume story, not a share-shift story. More modules shipped means more DSP units sold somewhere in the duopoly, regardless of which assembler ships the box. Read narrowly, the IPO is closer to a mild tailwind for the DSP layer as a whole than a threat to any part of it. The nuance is which half of the duopoly captures more of that incremental volume, and on the current product mix that favors Broadcom’s existing relationship with Innolight more than Marvell’s.
The risk that matters is a technology roadmap, not a funding round
Part of Innolight’s raise is earmarked for silicon photonics and next-generation module development, and Chinese module makers have been explicit about pursuing linear pluggable optics as a parallel track, an architecture that removes the DSP from the module entirely and shifts signal processing into the switch silicon. The stated purpose is reducing dependence on Broadcom and Marvell specifically. That is a real multi-year risk to the economics of the DSP layer, but it is a pre-existing roadmap risk that this listing accelerates with better funding rather than one it creates. It belongs on the list of things to track in Marvell’s data center commentary over the next several quarters, distinct from anything mechanical about the IPO itself.
Where the stock sits while this plays out
Marvell trades near 187, roughly 43 percent below its 52-week high of 329.88 and against a 52-week low of 61.44, a range that captures how violently sentiment on the optical and custom silicon story has swung this year. The stock carries a trailing price-to-earnings multiple in the mid-60s, and Street targets are split: KeyBanc raised its target to 400 from 385 in mid-July, while Erste Group moved the other direction with a downgrade to Hold the same week. The average target sits closer to 245 to 255, implying room to run without resolving which side of that analyst split is right. Innolight’s IPO does not change that setup. The LPO transition timeline is the actual variable that will.
