Zhongji Innolight (300308) Report Interpretation

Overview

UBS resumes coverage of Zhongji Innolight with a Buy rating, arguing that expanding AI data-center networking demand, leadership in 1.6T products and a developing 3.2T NPO opportunity can sustain strong growth despite concerns over AI capex, competition and US regulatory risk.

Core views

UBS's central thesis is that connectivity is becoming the bottleneck in AI infrastructure as computing power, memory bandwidth and GPU/ASIC cluster sizes rise. It expects bandwidth upgrades to begin with switch ASICs and then NICs, while network architectures move from one-layer switching toward two-layer fat-tree and potentially three-layer configurations. This raises optical-link content per GPU: UBS estimates the optical-transceiver attach rate per GPU will rise to 4-5:1 in 2028-30 from 2-2.5:1 in 2024-25. Global GPU/ASIC shipments are estimated to more than double to 28m units in 2027 from 13m in 2025. On this basis, UBS forecasts global datacom optical-transceiver revenue of roughly US$120-185bn in 2028-30, versus US$19bn in 2025, a c60% CAGR. It argues that networking capex may be the last part of AI infrastructure affected by any capex slowdown because connectivity remains a critical constraint on cluster scaling. For Innolight, UBS expects leadership in high-speed pluggable optical modules to persist. It cites faster time-to-market, with volume shipments of 400G, 800G and 1.6T products six months ahead of its closest peer; early SiPh development since 2017; active-alignment and thermal-management know-how; trusted customer relationships; and stronger production scale and supply-chain management. The report notes that 70% of high-speed product revenue came from SiPh in Q1 2026 and that Innolight's capacity and production are about 1.5x those of Eoptolink. UBS forecasts 800G shipments of 21m/22m units and 1.6T shipments of 31m/38m units in 2027/2028, implying market shares of approximately 25% and 40-45%, respectively. It also expects raw-material tightness to constrain sector shipments but believes Innolight may be prioritized in allocation because of its capacity and yield record. UBS rejects the view that near-packaged optics is merely a transitional technology before co-packaged optics. It expects pluggables, NPO and CPO to coexist over the next three to five years: NPO is positioned closer to the switch ASIC for higher bandwidth and lower power use, while retaining socket-based flexibility, easier serviceability and a more open ecosystem than CPO. UBS points to the March 2026 Open CPX MSA, NPO demonstrations at WAIC 2026 and industry commentary on NPO momentum as adoption evidence. It expects Innolight's SiPh+TFLN process expertise to let it capture higher content value as optical functions become more integrated. The company is expected to start 3.2T NPO shipments in 2027, initially for switches and later for xPUs, with 50k units in 2027 and 2.5m in 2028. UBS estimates 3.2T NPO will account for about 10% of sales by 2028 and 30% by 2030. Financially, UBS estimates revenue CAGR of 52% in 2026-29, driven by 800G/1.6T ramps, customized 2.4T products and 3.2T NPO. It forecasts revenue growth of 187% in 2026, 112% in 2027 and 42% in 2028, reaching Rmb329bn in 2028. Gross margin is expected to moderate from 2027 but remain above 40% by 2030, compared with 33.8% in 2024 and 42.0% in 2025. The report expects a better product mix to support blended ASPs, partly offset by early-ramp yield loss and raw-material costs. UBS has sharply raised estimates from November 2024, taking 2026E/2027E/2028E EPS to Rmb29.69/Rmb60.69/Rmb84.96, increases of 209%/479%/676% versus prior forecasts. It estimates 69% EPS CAGR in 2026-28, 16 percentage points above peers and 5 percentage points above Reuters consensus. UBS considers valuation attractive after the stock corrected 35% from its late-June peak on concerns over global AI capex, an FCC ban and competition. At the report date, the shares traded at 15x/10x 2027E/2028E P/E, which UBS describes as well below peers and representing a c50% discount. It applies a 25x 2027E P/E multiple, roughly one standard deviation above the historical average but below peers' c30x, to derive an Rmb1,500 target price. UBS argues that a higher multiple is warranted as Innolight moves beyond module assembly toward greater chip-design integration in the 3.2T era; the target implies a 0.9x PEG based on its 27% 2027-29E EPS CAGR. The major risk is a potential FCC ban on Chinese-made optical transceivers. The US represented c60% of Innolight sales in 2024-25, although UBS believes overseas manufacturing capacity and R&D centers can mitigate exposure, with overseas capacity now c85% of total capacity. Its scenario analysis characterizes a soft ban as high probability with estimated revenue impact below 5%, a medium ban as potentially reducing net profit by around 10%, and a low-probability hard ban as capable of causing revenue downside of up to 50%, net-profit downside of 50% or more, and gross margin falling to the teens. UBS also flags changing competition, failure to expand upstream while maintaining gross margin, and geopolitically driven supply constraints as downside risks.

Analysis framework

UBS starts with AI-networking demand drivers—GPU/ASIC volumes, bandwidth upgrades, network topology and optical-module attachment rates—to estimate the total addressable market. It then assesses Innolight's production, technology, customer and supply-chain advantages; models product ramps and margins; evaluates NPO versus CPO; applies scenario analysis to regulatory risk; and values the shares using a target P/E multiple and EPS growth comparison.

Methodology notes

• Industry AnalysisSupply-demand framework

  AI data-center optical-transceiver supply-demand and content-per-GPU analysis

  UBS links GPU/ASIC shipments, switch and NIC upgrades, and more complex network architectures to rising optical-transceiver demand and estimates market size, attachment rates and product mix.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  AI infrastructure connectivity value-chain analysis

  The report explains how improvements in compute and memory increase networking requirements, which in turn raises demand for high-speed optical modules and integrated optical technologies.

• Valuation methodsP/E and PEG Valuation

  Target P/E and PEG valuation

  UBS applies 25x 2027E P/E to estimated EPS of Rmb60.69 to derive the Rmb1,500 target price and compares the resulting 0.9x PEG with its EPS-growth forecast.

• Event-Driven and Behavioral FinanceEvent-driven analysis

  FCC-ban sensitivity scenarios

  UBS considers soft, medium and hard regulatory-ban outcomes and describes their potential effects on revenue, gross margin and net profit.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Zhongji Innolight (300308.SZ)

  Primary covered company and expected beneficiary of AI data-center optical-networking growth.

  Strengths

  Leadership in 800G/1.6T, production yield, supply-chain management, overseas capacity, SiPh expertise and expected 3.2T NPO leadership.

  Weaknesses

  New-product ramps may incur yield loss and higher costs; gross margin is expected to moderate from 2027.

  Comparison

  UBS expects Innolight to trade below peers despite estimated 69% 2026-28 EPS CAGR; its capacity and production are about 1.5x Eoptolink's.

  Risks

  Potential FCC restrictions, changing competition, supply constraints and failure to expand upstream while maintaining gross margin.

• Eoptolink

  Comparable and domestic optical-transceiver competitor.

  Weaknesses

  Innolight's capacity and production are described as about 1.5x Eoptolink's.

  Comparison

  Used as Innolight's closest capacity comparison.

• Tower Semiconductor

  Supplier partner supporting Innolight's SiPh PIC capacity.

  Strengths

  Described as the leading SiPh foundry supplier worldwide.

Key data

• Target priceRmb1,500Based on 25x 2027E P/E.
• Closing priceRmb921.00Price as of 12 Aug 2026.
• Forecast price appreciation62.9%UBS forecast return component over the next 12 months.
• Global datacom optical-transceiver TAMcUS$120-185bn in 2028-30Up c6-10x from US$19bn in 2025; c60% CAGR in 2025-30.
• GPU/ASIC shipments28m units in 2027UBS global tech estimate, more than double 13m in 2025.
• Innolight EPS CAGR69% in 2026-2816ppt above peers and 5ppt above Reuters consensus.
• Revenue forecastRmb329bn in 2028After estimated 187% growth in 2026, 112% in 2027 and 42% in 2028.
• 3.2T NPO revenue contributionc10% by 2028; 30% by 2030UBS expects shipments to begin in 2027 and ramp in 2028.
• Overseas manufacturing capacityc85% of totalCited as a mitigant to potential FCC-related risk.

Impact & implications

UBS views Innolight as a key beneficiary of global AI optical-networking demand. Its case rests on the company defending high-speed transceiver share while increasing content value through SiPh+TFLN and 3.2T NPO; however, this outlook remains sensitive to regulatory outcomes, supply availability and the pace of technology adoption.

Risks

• A potential FCC ban on Chinese optical transceivers could materially affect Innolight because the US accounted for c60% of sales in 2024-25.
• The competitive landscape could change as technology upgrades progress.
• Innolight may fail to expand upstream and maintain its current gross margin.
• Geopolitical tension and supply-chain constraints could limit shipment capability.
• A hard-ban scenario could lead to revenue downside of up to 50%, net-profit downside of 50% or more, and gross margin falling to the teens, according to UBS's sensitivity analysis.

What to watch

• The pace of 800G and 1.6T demand and shipment growth.
• Evidence of 3.2T NPO adoption, including Innolight's expected 2027 launch and 2028 ramp.
• AI infrastructure capex, especially networking investment and GPU/ASIC shipment growth.
• The FCC review and any restrictions on Chinese optical transceivers.
• Raw-material availability, production yields and gross-margin development.
• Competitive pricing and Innolight's ability to defend market share.