Rosenblatt: OpticalContext Behind Recent Sector We

Rosenblatt: Optical

Context Behind Recent Sector Weakness

> China InP Capacity Scare: Stocks traded poorly over the last 1–2 months due to fears that expanding indium phosphide (InP) capacity in China could crash laser pricing.

> CPO Timing Rumors: Reports suggested a potential one-year delay for large-scale Co-Packaged Optics (CPO) deployment compared to the base expectation of a 2H27 build and 2028 shipment.

> Short Seller Dynamics: Checks indicate short sellers are exploiting this mid-quarter noise without genuine conviction. Many plan to cover their positions before 2QCY26 earnings roll out in late July/early August and intend to go long on Optical stocks by 2027 to capture the 2028–2030 CPO ramp.

Assessing the China InP Threat

> Supply vs. Demand: Chinese players could add 20% to 40% to Rosenblatt's 2030 non-China InP laser capacity forecast of ~$26bn. However, total 2030 InP capacity is still projected to fall 40% behind aggregate demand, mitigating oversupply risks.

> Technological Moat: Near-term Western pricing remains secure because Chinese manufacturers currently lack 200G EMLs and high-power CW (continuous wave) lasers required for CPO and Near-Packaged Optics (NPO) applications. They currently possess low-power CWs for SiPo transceivers and 100G EMLs for 800G transceivers.

> Market Decoupling: Long-term Western and Chinese supply chain dynamics are unlikely to be fungible due to political decoupling and the technical shift toward CPO. Western leaders like Lumentum (LITE) and Coherent (COHR) are already fully sold out through 2027 and into 2028, bolstered by major pre-buys from companies like NVIDIA.

Dongshan Precision: Recently greenlit a $1.2bn expansion in Changzhou for InP lasers and 1.6T transceivers after acquiring Source Photonics.

Sanan IC: Pivoting from LEDs to InP, expanding capacity from 2,750 to 6,000 6-inch wafers per month.

HiSense Optoelectronics & Accelink: Also actively growing InP capacity

Architecture & Structural Industry Demands

> CPO Timeline Intact: The smaller-scale CPO timeline (2H26 build, 2027 ship) remains on track, with 2027 production estimates ranging from 30K to over 350K units each for NVIDIA and Broadcom. 2028 remains the definitive volume inflection year for large-scale CPO, with NPO filling the intermediate gap in 2027.

> Network Flattening Already Priced In: Concerns that technologies like Multipath Reliable Connection (MRC) and Resilient Network Graphs (RNG) will hurt demand are overblown. While larger radix Ethernet switches reduce transceivers per switch from 6–7 down to 3–4, exponential XPU volume growth easily offsets the lower ratio.

> Alternative Laser Architectures (GaAs / VCSELs): 200G VCSELs (made from Gallium Arsenide rather than InP) are emerging as intermediate NPO solutions. Broadcom is looking at them for 3.2T NPO platforms, while Lumentum and startup PicoJool have demonstrated VCSEL-based solutions to circumvent potential InP supply constraints in GPU-to-GPU interconnects.

Key Company Earnings Setups

> Lumentum ($Lumentum(LITE.US)): Toughest near-term setup because it has provided the most forward guidance, though it remains highly attractive at ~$700. Investors are watching if it can accelerate its CY27 target of $2bn in quarterly revenue with 40+% operating margins.

> Coherent ($Coherent Corp.(COHR.US)): Highly favored due to accelerating internal 200G EML/CW laser production and a major new ~$1bn/year 1.6T transceiver contract win with Google (sourcing Mitsubishi lasers). Long-term margins will be supported by 11-figure LTAs with NVIDIA for 2027–2030.

> Applied Optoelectronics ($Applied Optoelectronics(AAOI.US)): Expected to deliver a "beat and raise" watershed 2Q26. Driven by 800G sales to Amazon and Oracle, revenues are expected to exceed $200mn in 2Q26, with 3Q26 guidance pointing to ~$300mn and gross margin expansion well north of the ~30% level seen in 1H26.

> Arista Networks ($Arista Networks(ANET.US)): Positioned for strong growth via massive deployment wins with Meta, Microsoft, Oracle, OpenAI/Anthropic, and Google (which could become a 10% customer in 2026). High deferred revenue recognition is expected to fuel >30% revenue growth in 2H26 and 40+% in 2027.

> Fabrinet ($Fabrinet(FN.US)): Positioned for solid sequential Datacom growth with NVIDIA alongside initial revenue contributions from new Datacom customers Amazon and Lumentum.

> Viavi Solutions ($Viavi Solutions(VIAV.US)): Safe from InP noise as it has no direct exposure. Growth is driven by testing systems (transceiver/OCS) and MilAero products, with CPO/NPO testing revenues slated to begin this fall.

> Ciena ($Ciena(CIEN.US)): Ciena's order book suggest that Data Center Interconnect (DCI) and scale-out orders are poised to reaccelerate from previous levels.