In the game of optical interconnect, which U.S. stocks benefit and which face pressure?

ComputeX 2026 has completely rewritten the narrative of CPO (Co-Packaged Optics) – Jensen Huang announced the silicon photonics roadmap and invested $2 billion each in Coherent and Lumentum in March 2026 to lock in laser supply; Marvell made the call: the next bottleneck for AI is not computing power, it's "connectivity." With the trends covered, today we only talk about the most practical aspects: which U.S. stocks benefit, what's the logic, and whose position is actually not that solid.

First, a measuring stick: the closer to "optical chips/lasers" and "advanced packaging," the deeper the moat; the closer to "pure module assembly," the more one needs to be wary of being internalized by CPO in the long run. When looking at targets, first ask which end of the stick it stands on.

① Optical Module/Transceiver Leaders: Most Direct, Also the "Hottest"

Coherent (COHR) is the transceiver shipment leader, with about 25% market share. FY2025 revenue was $5.8 billion (+23%), data center book-to-bill ratio exceeded 4x, and indium phosphide laser orders are already booked until 2027. The stock price has risen about 449% over the past year. Applied Optoelectronics (AAOI) is a high-beta player – Q1 2026 revenue +51%, data center revenue doubled, but it's still operating at a loss with a beta as high as 3.76, indicating extreme volatility. This segment benefits most directly from the volume ramp of 800G/1.6T, but valuations and stock prices are already extremely stretched, and CPO internalization is a long-term sword of Damocles – as optical engines are packaged next to the chip, the value of pluggable modules will partially shift.

② Lasers/Optical Chips: The Most Critical and Constrained "Achilles' Heel"

Whether pluggable, LPO, or CPO, all require the "light-emitting heart" – EML lasers and CW light sources. Lumentum (LITE) holds the most scarce position here – with about 50-60% share in high-end EML, it's currently the only player globally mass-producing 200G/lane EML (a key component for 1.6T modules). Latest quarterly revenue was $666 million, up 65% YoY, with guidance for next quarter's growth to surge to about 90%. It was added to the S&P 500 in March, and its stock price has risen over 10x in the past year. McKinsey estimates that 800G capacity will still be 40-60% short by 2027. The logic for this segment is the purest: surging demand, rigid supply. The risk is valuation has reached about 28x TTM revenue, leaving no room for error, and scale-up CPO first shipments won't happen until late 2027.

③ Switch Chips/CPO Platforms: The True System Players

This is the segment that directly packages the optical engine next to the switch chip. NVIDIA (NVDA) sets the pace with its Quantum-X/Spectrum-X photonic switching platforms and the Rubin/Feynman roadmap; Broadcom (AVGO) takes a modular approach, having reached the third-generation Tomahawk 6 Davisson CPO and is developing the fourth generation; Marvell (MRVL) entered via the acquisition of Celestial AI, guiding its CPO business to reach $500 million by FY28 Q4 and double to $1 billion the following year. This tier is the "problem-setter," with high certainty, but CPO is just one of many businesses for them, requiring a look at overall valuation.

④ Foundry/Advanced Packaging: The "Decisive Battlefield" for CPO

If you agree that "the difficulty of CPO lies in packaging," this segment deserves the most attention but is most easily overlooked. TSMC's (TSM, U.S. ADR) COUPE silicon photonics platform enters mass production in 2026 and is the common foundry for NVIDIA and Broadcom's CPO, using SoIC-X to directly stack electrical chips onto optical chips; its SoIC monthly capacity is rumored to surge to 30k-40k wafers by end-2026. Additionally, Corning (GLW) supplies the specialty fibers and connectivity solutions required for CPO. The investment implication is simple: even if you can't accurately judge which chipmaker or module maker will win, betting on the "packaging/foundry" process, which is a necessary step, is a more robust "pick-and-shovel" approach – no matter who wins, packaging is indispensable.

Finally, Three Buckets of Cold Water

First, penetration is extremely low – CPO's share in optical networks will still be less than 1% in 2026, with TrendForce expecting it to exceed 35% only by 2030. The story is far from realized, and pluggables remain the mainstay. Second, valuations and stock prices are already extremely stretched – COHR up ~449% in a year, LITE up over 10x, AAOI up over 10x. Customer concentration is high among a few cloud vendors, amplifying the impact of any "missed expectations." Third, demand concerns – if cloud vendor capex growth marginally slows, coupled with engineering variables like 6-inch InP yield and LPO delaying CPO, the high-growth narrative could lose support. The certainty of the trend is extremely high, but this is a marathon driven by slow-moving variables.