2026 China Optical Transceiver Industry: Market Size and Competitive Landscape Research Report

Abstract

Optical transceivers are the "capillaries" of the global communications network—every piece of data traversing a network passes through one to complete the conversion between electrical and optical signals. In 2024, those capillaries experienced their most explosive expansion on record: the global optical transceiver market reached approximately USD 17–18 billion, growing roughly 52% year-on-year, marking the highest growth rate in nearly a decade. The driver was not voice calls or video streaming—it was AI, specifically the GPU cluster interconnects required by hyperscale data centers worldwide to support large-model training and inference.

Every AI server needs several optical transceivers to connect to its neighbors; a training cluster housing ten thousand GPUs can easily consume tens of thousands of 800G modules. The four hyperscale cloud vendors—Microsoft, Google, Meta, and Amazon—collectively exceeded USD 300 billion in AI capital expenditure in 2025, a significant step up from approximately USD 200 billion in 2024, with 2026 capex projected to surpass USD 600 billion (+70%+ year-on-year). NVIDIA's Blackwell B200/GB200 entered full-scale volume production in 2025, the B300 (Blackwell Ultra) ramped in the same year, and the Rubin architecture is set to debut in late 2026 to early 2027. This continuous GPU generation cadence keeps hyperscale data-center buildout at extreme intensity, and every percentage point of AI capex leaves a clear imprint on transceiver demand.

Chinese vendors are the greatest beneficiaries. Seven of the global top-ten optical transceiver suppliers are Chinese. Innolight (300308) leads with approximately 23.4% global market share—retaining the No. 1 position for three consecutive years—with FY2025 revenue of CNY 38.24 billion (+60.25%) and net profit of CNY 10.797 billion (+108.78%), crossing the CNY 10 billion net-profit threshold for the first time. Eoptolink (300502) posted FY2025 revenue of CNY 24.842 billion (+187.29%) and net profit of CNY 9.532 billion (+235.89%), among the highest growth rates in the A-share technology sector. HGtech Photonics (002281) FY2025 revenue: CNY 11.929 billion (+44.20%). Accelink (300394) FY2025 revenue: CNY 5.163 billion (+58.79%) and net profit CNY 2.017 billion (+50.15%). The combined global share of the CR3—Innolight + Eoptolink + HGtech Zhongyuan—now exceeds 55%, further consolidating from 2024's 50%+. This is rare in Chinese manufacturing: not "ranked in the top three" but "accounting for more than half of the world."

Speed-tier migration is another major theme. 400G transceivers have entered the mature-price-decline phase in 2025. 800G global shipments in 2025 surpassed 15 million units (+60%+ year-on-year, per Cignal AI), becoming the dominant product in data-center deployments. 1.6T has entered an all-out mass-production race: Innolight shipped approximately 3 million units of 1.6T in FY2025, capturing 50%+ global share, while locking up roughly 70% of Lumentum's 12-million-unit EML chip capacity; Eoptolink is one of the world's first manufacturers to achieve large-scale mass production and delivery of 1.6T, scaling up from Q3 2025. Looking ahead, 3.2T will not take shape until after 2027, but the technical roadmap is clear: co-packaged optics (CPO) will be the mainstream architecture for next-generation switches, with TSMC beginning CPO module production in 2025.

Yet beneath the brilliance lie hidden dangers. Upstream optical chips—especially high-end EML lasers and indium phosphide (InP) substrates—remain the deepest weak link in China's optical communications supply chain. The InP substrate domestic substitution rate is below 15%; the rate for high-end EML lasers needed for 1.6T is below 25%; and the supply-chain risk posed by DSP chips (Marvell/Broadcom) looms as a latent threat. However, upstream breakthroughs are emerging: YJSemi (688498) posted FY2025 revenue of CNY 601 million (+138.5%), with data-center CW laser shipments exceeding 1 million units and 100G EML passing customer validation; Changelight (688048) has commenced volume production of 100G EML with 200G EML entering customer sampling. The next battle for domestic substitution in optical transceivers will not be fought in module factories—it will be fought in chip foundries.

This report takes 2026 as its observation baseline. It systematically reviews the market size, supply-chain structure, competitive landscape, sub-segment tracks, technology evolution paths, industrial cluster distribution, risks, and five-year forecasts for China's optical communications transceiver industry, and positions optical transceivers within China's technology manufacturing chain from the "AI compute quadruplet" perspective (PCB × lasers × AI servers × optical transceivers).

Core findings:

• "The most certain Chinese advantage segment in the AI compute downstream chain": Unlike semiconductors (subject to sanctions / technology gaps) or semiconductor equipment (heavy import dependency) or AI servers (assembly character), optical transceivers are the node at which China has genuinely achieved dominant global market share in the AI infrastructure supply chain.
• CR3 global > 55% expected to hold: Innolight + Eoptolink + HGtech Zhongyuan's combined global share now exceeds 55%; their capacity expansion and product-iteration head starts are difficult to close in the short term.
• Upstream optical chips are the next battlefield: The competitiveness of domestic module manufacturers has already been validated. Over the next five years, YJSemi (688498) and Changelight (688048) carry both the pressure and opportunity of domestic substitution.
• Price wars and AI bubble are the twin risks: Intense internal competition in China drives accelerating ASP decline; if AI capex falls short of expectations, 800G/1.6T demand could soften in phases.
• CPO is a distant concern, not an immediate threat: Large-scale replacement of pluggable modules by CPO is not expected until after 2027–2028; until then, the market window for pluggable transceivers remains wide open.

Key data at a glance (updated to FY2025):

• Global optical transceivers 2025E: approx. USD 23–26 billion; 2030E: approx. USD 40–50 billion (CAGR approx. 18%–22%).
• Chinese vendors hold 7 of 10 global top positions; CR3 global share >55%.
• Innolight (300308) FY2025 revenue CNY 38.24 billion (+60.25%); net profit CNY 10.797 billion (+108.78%).
• Eoptolink (300502) FY2025 revenue CNY 24.842 billion (+187.29%); net profit CNY 9.532 billion (+235.89%).
• HGtech Photonics (002281) FY2025 revenue CNY 11.929 billion (+44.20%); Accelink (300394) FY2025 revenue CNY 5.163 billion (+58.79%).
• YJSemi (688498) FY2025 revenue CNY 601 million (+138.5%); net profit CNY 191 million, turned profitable.
• 800G global shipments 2025E: over 15 million units (+60%+); 1.6T 2025: Innolight ~3 million units; 2026 global demand projected to exceed 10 million units.
• Four hyperscale cloud vendors' 2025 AI capex: over USD 300 billion; 2026 forecast: over USD 600 billion.
• InP substrate domestic substitution rate <15%; high-end EML laser domestic substitution rate <25%.

Chapter 1　Definitions, Classification, and Full Industry-Chain Overview

1.1　The Nature of Optical Transceivers: The "Translator" of Optical Networking

Modern data communications networks rely on optical fiber for long-distance data transmission—light travels through glass fiber with extremely low loss, theoretically carrying far more information per second than any copper wire. But computers, servers, and switches process only electrical signals internally. This creates a fundamental mismatch: computation lives in the electrical domain, transmission lives in the optical domain.

An optical transceiver (also called an optical transceiver module) is the solution to this mismatch. It is a precision optoelectronic integrated device incorporating lasers, photodetectors, driver chips, signal-processing chips, optical components, and electrical interfaces. Its function in a single sentence: on the transmit side, convert electrical signals output by chips into laser pulses and couple them into optical fiber; on the receive side, recover electrical signals from laser pulses arriving from the fiber and return them to the chip. One transmit, one receive—forming a complete optical-electrical "translation" channel.

Optical transceivers may sound simple, but their manufacturing difficulty is extreme. Within a tiny physical space (typical QSFP-DD dimensions roughly 35 mm × 13 mm × 9 mm), they integrate lasers (wavelength precision to 0.01 nm), detectors, driver chips, DSP chips, optical coupling lenses, isolators, multiplexer/demultiplexer components, and high-speed electrical interfaces. At 800G rates, the internal throughput of a single module is equivalent to simultaneously transmitting 80 billion bits per second—at this speed, any minute temperature drift, mechanical vibration, or manufacturing deviation can spike the bit error rate.

Optical transceivers are the fundamental building blocks of the entire optical communications network. From server connections in enterprise campus networks to inter-data-center long-haul transmission, from carrier backbone spans of tens of thousands of kilometers to GPU-to-GPU high-speed interconnects in AI compute clusters—every link requires optical transceivers. An ultrascale compute cluster housing 100,000 H100/H200 GPUs can easily consume more than one million high-speed transceivers.

1.1.1　Internal Structure

A standard optical transceiver comprises the following core functional blocks:

Laser (transmit side): The source that converts electrical signals into optical signals. Common types include DFB (distributed feedback laser), EML (electroabsorption modulated laser), VCSEL (vertical-cavity surface-emitting laser), and CW (continuous-wave laser, used in silicon-photonics solutions). DFB suits 10G–25G rates; EML is the mainstream choice for 100G and above; VCSEL offers lower cost for short-reach multimode datacenter scenarios; CW lasers are paired with silicon-photonic modulators.

Photodetector (receive side): Converts incoming optical signals from fiber back into electrical signals. Common types are PIN photodiodes (lower-speed scenarios) and APD avalanche photodiodes (high-sensitivity, long-reach scenarios).

Driver IC: Provides high-speed modulation current to the laser. In LPO (linear pluggable optics) solutions, the driver IC accepts linear electrical signals directly; in conventional solutions, a CDR (clock-data recovery) or DSP chip preconditions signals before the driver.

DSP chip: Digital signal processor responsible for forward error correction (FEC), signal equalization, and dispersion compensation. The core chip in 800G/1.6T transceivers, supplied primarily by Marvell (MRVL) and Broadcom (AVGO), with high power but strong link tolerance.

Optical sub-assembly (OSA): Includes collimating lenses, isolators, and wavelength-division multiplexing/demultiplexing components. Responsible for combining multiple laser signals into a single fiber, or separating multiple wavelengths from a single fiber to individual detectors.

Packaging and housing: Manages precision optical alignment, thermal control, and electrical interconnects. High-speed transceivers have substantial thermal dissipation; an 800G module typically dissipates 14–20 W, imposing stringent thermal design requirements.

1.1.2　Optical Path Solution Comparison

Solution	Characteristics	Typical Speed	Representative Vendors
Direct modulation/direct detect (IM-DD)	Low cost, simple structure	100G/400G (short reach)	Most datacom module manufacturers
EML external modulation	High optical power, long reach	100G/400G/800G (medium-long reach)	Innolight, HGtech Photonics, Coherent
Coherent	Ultra-long reach (100km+), high spectral efficiency	100G/400G/600G (telecom backbone)	Coherent, Lumentum, HGtech Photonics
Silicon photonics (SiPh)	High chip integration, CMOS-compatible	400G/800G/1.6T (ramping)	Intel, TSMC ISLD, YJSemi, ZTE Micro
CPO (co-packaged optics)	Co-packaged with ASIC, lowest power	1.6T/3.2T (early commercial)	TSMC, Intel, Marvell

1.2　Classification System

1.2.1　By Application

Datacom (DC): High-speed interconnects within and between data centers. Customers are cloud companies (Meta, Google, Microsoft, Amazon, ByteDance, Tencent, Alibaba) and AI infrastructure builders. In 2024, datacom accounted for approximately 50%+ of the global transceiver market and is the fastest-growing sub-segment. Datacom modules prioritize low power, low cost, and high density; they use PAM4 modulation and span 100 m to 10 km.

Telecom: Used in carrier networks—fronthaul, midhaul, and backhaul—as well as DWDM backbone systems and FTTH access networks. In 2024, telecom accounted for approximately 35%. Telecom modules require wider temperature ranges (−40°C to +85°C), longer reach (10 km to thousands of km), and higher long-term reliability, typically using coherent technology (backbone) or CWDM/DWDM (access).

Access/Passive Optical Network (PON): GPON/XGS-PON fiber-to-the-home scenarios; relatively low rates (1G–10G), large volumes, low unit prices. Approximately 15% of the market.

1.2.2　By Speed

Speed is the most critical classification dimension and the main axis of product iteration:

Speed	Peak Year	Status (2024/2025)	Modulation	Key Demand
10G	2010s	Mature/low price	NRZ	Legacy DC, access
25G	2015s	Mature	NRZ	DC server access, 5G fronthaul
100G	2018–2020	Mainstream/declining price	NRZ/PAM4	DC standard, telecom backbone
200G	2020–2021	Transitional	PAM4	Brief transition
400G	2022–2025	Highest shipment volume	PAM4	DC/AI cluster mainstream
800G	2024–2026	Rapid ramp (+6× YoY)	PAM4/100GBaud	AI supercompute clusters
1.6T	2025–2027	Early mass production (China ~45%)	200GBaud PAM4/coherent	Next-gen AI clusters
3.2T	2027+	R&D phase	High-order PAM/coherent	Ultra-large inference

1.2.3　By Form Factor

QSFP28: 100G mainstream, backward-compatible, high port density.

QSFP-DD (Double Density): Main form factor for 400G/800G, doubling the channel count of QSFP28 to 8 high-speed differential lanes.

OSFP (Octal SFP): Second major 800G form factor; superior thermal design; broadly adopted in NVIDIA InfiniBand ecosystems.

CPO (Co-Packaged Optics): Integrates the optical engine on the same substrate as the switch ASIC; eliminates pluggable interfaces, sharply reducing power and signal losses.

LPO (Linear Pluggable Optics): A technology approach rather than a form factor; eliminates DSP, reducing power and latency. Discussed as a 1.6T transition option between DSP-based and CPO solutions.

AOC (Active Optical Cable): Integrates two optical transceivers into the cable ends; suited for short-reach (0.5 m–100 m) rack-to-rack interconnects.

DAC (Direct Attach Copper): Copper passive cable for extreme short-reach (≤7 m) cost-sensitive scenarios.

1.2.4　By Reach

Reach Class	Typical Range	Application	Example Products
SR (Short Reach)	≤100 m (multimode)	Same-room/floor DC	400G SR4, 800G SR8
DR	500 m–2 km	DC campus interconnect	400G DR4/FR4
LR/ER	10 km–40 km	Metro DC interconnect	400G LR4, 800G LR8
ZR/ZR+	80 km–2,000 km	Carrier backbone, metro	400G ZR (coherent), 600G ZR+
Long-haul coherent	1,000 km–10,000 km	Telecom backbone, submarine	100G/200G DWDM coherent

1.3　Full Supply-Chain Overview

The optical transceiver supply chain spans three tiers: upstream raw materials and core chips, midstream optical components and module manufacturing, and downstream network equipment and application systems.

1.3.1　Upstream: Core Materials and Chips

InP substrate, EML/DFB laser chips, DSP chips, and detector chips are the highest-barrier, lowest-domestic-substitution elements in the supply chain.

Substrate materials: InP (indium phosphide) is the most critical substrate material for high-speed laser chips (EML/DFB); global capacity is dominated by Coherent (II-VI), Sumitomo Electric (Japan), and Wafer Technology (UK); China's domestic substitution rate is below 15%.

Laser chips: DFB, EML, VCSEL, and CW. YJSemi (688498) has achieved large-scale domestic production of 25G DFB chips. Changelight (688048) has put 100G EML chips into production. But the domestic substitution rate for high-end EML chips for 1.6T remains below 20%.

DSP chips: Marvell (MRVL) and Broadcom (AVGO) jointly dominate global 400G/800G PAM4 DSP supply (~70%–80% combined share)—the clearest single "chokepoint" risk in China's optical transceiver supply chain.

AWG wavelength-division multiplexing components: Accelink's investee Sivers Photonics has broken the foreign monopoly; domestic substitution rates for 100G/400G AWG are relatively high.

1.3.2　Midstream: Optical Components and Module Assembly

Optical components (OSA assembly) are high-margin sub-assemblies produced from laser chips, detectors, lenses, and isolators. Accelink Technology (300394) is China's leading passive optical component manufacturer.

Module assembly—integrating OSA, driver chips, DSP chips, and housings into pluggable finished goods—is the stage where Chinese manufacturers are globally strongest.

Optical fiber cables are an independent supply chain but tightly coupled with transceivers. Yangtze Optical Fibre and Cable (61368) is a global top-three manufacturer.

1.3.3　Downstream: Network Equipment and Application Systems

AI servers and GPU clusters, Ethernet switches (Broadcom Tomahawk ASIC-based), InfiniBand switches (NVIDIA Quantum series), 5G base-station network equipment, ROADM DWDM systems, and FTTH OLT terminals are the principal downstream consumers.

1.4　"AI Compute Quadruplet" Relationship

Optical transceivers, together with PCBs, lasers (optical chips upstream), and AI servers, form the "AI compute quadruplet"—each mutually reinforcing and jointly constituting China's technology-manufacturing strategic opportunity in the AI era.

1.5　Measurement Standards and Certification Ecosystem

Achieving hyperscale data center customer qualification (12–18 months; MSA compliance, IEEE 802.3 standards, CMIS management interface, customer private specifications, factory audits) represents the most effective market-entry barrier for new entrants.

1.6　Technical Standards and Wavelength Planning

Chinese modules operate primarily in the O-band (1310 nm) for short/medium datacom and the C-band (1550 nm) for long-haul coherent telecom. CWDM (20 nm spacing) handles medium-reach modules like 400G FR4/800G FR8; DWDM (100 GHz or 50 GHz grid) supports carrier backbone systems. Single-mode fiber (SMF, 9 µm core) serves all 400G+ datacom modules; multimode fiber (MMF, 50–62.5 µm core) remains for short-reach (≤300 m) VCSEL-based SR modules.

1.7　International Optical Communications Community

Key events: OFC (Optical Fiber Communication Conference, March, US—global premier conference; 2024 OFC featured 1.6T LPO prototype demonstrations), ECOC (European Conference on Optical Communications, September—more research-oriented), CIOE (China International Optoelectronic Exposition, September, Shenzhen—China's largest annual optical communications trade show).

Chapter 2　Global Competitive Landscape and Major Overseas Vendors

2.1　Global Competitive Landscape Overview

The 2024 global optical transceiver market of roughly USD 17–18 billion shows a historic structural shift: Chinese vendors dominate the top ten, and China holds more than 50% of total market share.

• China: Combined share exceeds 50% (Innolight alone ~22%).
• Western vendors (Coherent, Lumentum): roughly 15%–20% in telecom coherent and high-end DSP modules, but ceding ground in 400G/800G datacom.
• Japan/others (Sumitomo, Fujitsu): focused on telecom-side components, share gradually eroded.

In the 800G sub-segment, Coherent and Lumentum together hold roughly 40% of Western-vendor share but only ~15%–20% of the global total, far below the Chinese combined share of over 60%.

2.1.1　Tiered Structure

Tier 1 (global share ≥15%): Innolight (~22%)

Tier 2 (5%–15%): Eoptolink (8%–10%), Coherent (10%), HGtech Zhongyuan (7%–8%), HGtech Photonics (6%–8%)

Tier 3 (2%–5%): Lumentum, Accelink Technology (components), Cambridge Industries Group, Broadex, Taitwah Communications, Mingpu Optoelectronics

Regional/specialist vendors: Sumitomo Electric, Fujitsu (telecom side); YJSemi (chips, not modules); Changelight (chips, not modules)

2.2　Coherent Corp (COHR)—Western Integrated Optical Communications Leader

Coherent Corp was formed when II-VI Incorporated acquired Finisar (~USD 5.8 billion, 2022) and then merged with Coherent (laser/photonics specialist) under the II-VI name, which was then rebranded Coherent Corp. The company spans semiconductor materials (InP/GaAs), lasers, transceivers, and industrial lasers.

FY2024 financials: Total revenue ~USD 4.7 billion; communications segment (transceivers + optical components) ~USD 2.7–3.0 billion.

Product lines: One of the few companies globally with a vertically integrated InP substrate → EML laser chip → optical transceiver chain. Its 800G coherent modules (800ZR/ZR+) lead in telecom carrier markets.

Relevance to China: Coherent is an important supplier of InP substrates and EML laser chips to Chinese transceiver manufacturers; export control expansion to these materials poses an indirect risk to the Chinese supply chain.

2.3　Lumentum Holdings (LITE)—Laser and Telecom Photonics Leader

Spun off from JDSU, Lumentum (NASDAQ: LITE) focuses on photonics for telecom and datacom.

FY2024 financials: Total revenue USD 1.5–1.8 billion; telecom/datacom photonics components > 70% of revenue. High gross margins (38%–42%).

EML chips: Lumentum (via Oclaro acquisition) is one of the world's largest EML chip suppliers; many Chinese transceiver manufacturers are customers.

AI datacom opportunity: Lumentum explicitly positioned AI datacom as its core growth track in 2024—high-power EML and CW laser components for AI supercompute clusters command premium margins.

2.4　Marvell Technology (MRVL)—Critical DSP Chip Supplier

Marvell is essentially the provider of the most expensive "invisible component" in optical transceivers—the DSP chip.

Business: Marvell's PAM4 DSP series (Spica, Alaska) are the mainstream signal processors for global 400G/800G pluggable transceivers, handling FEC, equalization, dispersion compensation. Combined with Broadcom, the two share ~70%–80% of global DSP market.

AI/DC revenue (FY2024): ~USD 1.5 billion from AI/data center, growing strongly.

China impact: Marvell DSPs are used throughout Chinese transceiver manufacturers' 400G/800G product lines. Any escalation of US export controls to cover high-end PAM4 DSP chips would directly constrain Chinese module production—hence the urgency behind LPO technology development.

2.5　Broadcom (AVGO)—ASIC and DSP Dual Positioning

Broadcom's Tomahawk switch ASIC series sets the pace for data center switch port speeds, directly determining which optical transceiver speeds are in demand. Broadcom Tomahawk 5 (51.2 T) drives 800G deployment; Tomahawk 6 (expected 2025, 102.4 T) will drive 1.6T/3.2T. Additionally, through its Inphi acquisition (2021, ~USD 10 billion), Broadcom entered DSP and photonic integration, competing with Marvell.

2.6　Japanese Vendors: Sumitomo Electric and Fujitsu

Sumitomo Electric: Second-largest global InP substrate supplier; strong in fiber amplifier components; serves Japan domestic and East Asian telecom carriers.

Fujitsu: Historical strength in ultra-high-speed coherent DWDM systems; global transceiver share has sharply contracted; primarily serves domestic Japanese carriers.

2.7　Western Vendor Strategy vs. Chinese Response

Western vendors' strategy: "defend technological moats"—avoid head-on competition in pluggable datacom modules where Chinese manufacturers hold cost/scale advantages; instead focus on high-end coherent modules (telecom backbone, submarine cables), vertical material chains (InP, EML, DSP), and CPO.

Chinese vendors' response: short-term—drive 400G/800G/1.6T pluggable modules to global cost floor, highest yield, fastest delivery; long-term—independently break through EML laser and InP substrate (YJSemi, Changelight), advance LPO to reduce reliance on Marvell DSP.

2.8　Historical Evolution of the Global Market and China's Rise Path

From the early 2010s (when Western and Japanese vendors held 60%+ of the global market) through the 100G breakthrough by Chinese companies (2015–2018), the 400G first-mover moment (2019–2021), to the 800G explosion confirming "China's 7 of 10" (2022–2024)—this represents a remarkable fifteen-year power transfer.

2.9　Chinese Vendors' Competitive Advantage System

Speed: Existing hyperscale customer certifications allow rapid entry into new speed tiers (800G → 1.6T) without restarting 12–18-month qualification cycles. Cost: Scale manufacturing in Suzhou (mature electronics ecosystem, local maintenance for JUKI/ASM placement equipment, precision mechanical sub-suppliers). Talent: Returnee silicon-valley engineering teams + deep domestic university pipeline (UESTC, Wuhan University, Fudan University). Customer relationships: Strategic supplier status with Meta/Google/Microsoft, enabling joint development of interface specs across speed transitions.

2.10　Finisar (II-VI) Merger History and Western Consolidation

The Finisar → II-VI → Coherent Corp consolidation was intended to create a vertically integrated Western champion. The reality: cultural integration friction, limited progress in 800G datacom volume versus Chinese rivals, and ongoing financial pressure suggest that acquisitions alone cannot replicate a decade of Chinese precision manufacturing accumulation.

2.11　Source Photonics: Acquired to Fill Telecom Gaps

Source Photonics, once a global top-10 transceiver supplier, was acquired by Innolight in 2019. This gave Innolight access to North American carrier customer networks and certifications, rapidly filling a gap in its telecom-side transceiver portfolio.

2.12　Channel and Distribution Structure

Direct sales: Major vendors supply hyperscale data center customers under multi-year framework agreements. Distribution: Medium-enterprise and regional carriers served via distributors (Ingram Micro, TD SYNNEX). ODM/OEM: Some Chinese manufacturers produce private-label modules for major equipment brands. E-commerce/B2B platforms: Active secondary market for 10G–100G products.

2.13　Technology Route Divergence and Market Segment Competition

EML+DSP conventional, LPO linear drive, silicon photonics + CW, CPO co-packaged, and coherent (ultra-long reach) are the five main technology approaches. Chinese vendors' multi-track parallel investment (especially Eoptolink's SiPh and LPO bet) reduces single-route failure risk and is the strategic insurance policy for the 2025–2028 transition.

Chapter 3　Macro-Environment Analysis (PEST)

3.1　Political/Regulatory Environment

3.1.1　China Industrial Policy: Comprehensive Support

"East-Data-West-Computing" national strategy (launched 2021); "14th Five-Year" Information and Communications Development Plan; "15th Five-Year" Guidelines (Nov 2025); MII "shortfall" projects for EML/DFB optical chips and InP substrates; 2025 national compute capacity target of 300 EFLOPS.

3.1.2　US Export Controls: A Suspended Sword

BIS rules (Oct 2022, Oct 2023) have targeted advanced AI compute chips (A100/H100) and semiconductor manufacturing equipment. DSP chips (Marvell/Broadcom PAM4 series) and optical transceiver finished goods remain outside the control list as of 2024, but the policy trajectory is toward broader coverage. InP substrates and high-end EML chips from US suppliers (Coherent, Lumentum) face latent exclusion risk if controls expand.

3.1.3　Supply-Chain Diversification Pressure

US chip legislation and allied supply-chain localization programs exert indirect pressure on optical chip production; replicating China's transceiver module manufacturing ecosystem (engineering talent, cost structure, supplier density) in alternative regions is practically infeasible in the near term.

3.2　Economic Environment

3.2.1　AI Capex Explosion

2024 AI infrastructure capital expenditure by major hyperscalers: Meta USD 37–40 billion (+33%), Microsoft ~USD 50 billion, Google/Alphabet ~USD 52 billion (+52%), Amazon AWS ~USD 49–53 billion, ByteDance ~RMB 50 billion (USD 7 billion). Total hyperscale AI capex exceeded USD 200 billion, with 2025 growth expected at 25%–40%. An NVIDIA DGX GB200 NVL72 rack requires 72+ 800G modules; a 10,000-GPU training cluster consumes 5,000–100,000 high-speed transceivers.

3.2.2　ASP Curves and Pricing Pressure

400G ASP fell from ~USD 300 (2020) to ~USD 80–120 (2024). 800G is on the same trajectory: ~USD 200–250 (2024) → ~USD 130–180 (2025E). Falling ASP benefits the largest, highest-yield manufacturers by widening cost gaps over smaller rivals.

3.2.3　5G Inventory and Stable Telecom Demand

Post-peak 5G construction (2021–2023), 2024 sees continued deep-coverage buildout. Fronthaul (25G SFP28), mid/backhaul (100G–200G) demand remains steady for HGtech Photonics, Accelink, and Cambridge Industries.

3.2.4　Exchange Rate and Supply-Chain Costs

USD/RMB ~7.1–7.3 in 2024, favoring export-oriented transceiver manufacturers. Upstream components (InP, EML, DSP) are USD-denominated; exchange-rate impacts on imports are roughly neutral.

3.3　Social and Technological Environment

3.3.1　Large AI Models: "Optical Transceiver Consumable" Attribute

Each generation of large models demands roughly 10× more compute than its predecessor. Inference-side demand is growing fastest and is continuous—every ChatGPT, Claude, or Gemini query triggers a GPU inference cluster, creating sustained recurring replacement demand for optical transceivers as consumables.

3.3.2　Accelerating Technology Iteration

Historical pace: one speed generation every 4–5 years (10G → 100G → 400G). AI demand has compressed this to roughly 2 years (400G → 800G → 1.6T), accelerating attrition of mid-sized players and reinforcing head-count concentration. Each speed transition's first-mover advantage further compounds Chinese market share.

3.3.3　CPO and Silicon Photonics Technology Variables

CPO: TSMC 3DFabric CPO modules entered production in 2025 (delivering samples to Broadcom and NVIDIA); power savings 30%+. Large-scale CPO penetration impacting pluggable market is estimated for 2027–2028—2026 penetration stays under 5%. SiPh: Already ~15%–25% of 400G/800G DR4/FR4 solutions; rising to 30%–40% in 1.6T by 2027–2029. TFLN: New modulator material with >100 GHz bandwidth; Eoptolink using TFLN for 1.6T modules.

3.4　Comprehensive PEST Assessment

Dimension	Direction	Strength
Politics—China policy support	Positive (compute infrastructure, domestic substitution)	★★★★★
Politics—US export controls	Negative (DSP/EML/InP latent supply risk)	★★★☆☆ (not yet directly targeted)
Economics—AI capex explosion	Positive (strongest demand driver ever recorded)	★★★★★
Economics—ASP decline pressure	Neutral/negative (scale leaders benefit; SMEs at risk)	★★★☆☆
Social—LLM "continuous consumption"	Positive (demand is durable)	★★★★☆
Technology—CPO substitution risk	Negative (long-term) / limited before 2026	★★☆☆☆
Technology—speed iteration acceleration	Positive (head leaders solidify)	★★★★☆

Overall: extremely high-prosperity cycle in 2024–2026; leading Chinese transceiver manufacturers are at a historically optimal operating window; primary risks are upstream supply-chain dependency and medium-term CPO technology transition.

3.5　The Boundary: Where Does US-China Tech Competition Affect Optical Transceivers?

Direct controls (GPU/HBM/EDA/lithography) are already in place and indirectly affect domestic compute expansion speed. Controls not yet in force (DSP chips, EML chips, InP substrates) carry escalation risk. Chinese finished transceiver exports remain unrestricted today, protected by the mutual economic deterrence—any US import restriction would harm American hyperscalers' ability to build AI infrastructure. "Technology mutual lock-in" is the strongest near-term shield.

3.6　Domestic DC Policy: Compute Corridors and Green Computing

Compute corridor plan: MII is building long-haul optical backbone links (500–1,000 km hops) connecting eastern compute-dense zones (Beijing, Shanghai, Guangshen) to low-electricity-cost western zones (Inner Mongolia, Ningxia, Guizhou). This drives large volumes of 100G/200G/400G telecom coherent modules—incremental opportunity for HGtech Photonics and Fiberhome Communications.

Green DC policy: NDRC PUE ≤1.3 for new large data centers. LPO and CPO modules' power advantages directly align with this policy. Liquid-cooling adoption (PUE ≤1.1) provides the infrastructure environment for CPO's high thermal density.

3.7　Industry Associations and Standards Organizations

CODA (China Optical & Electronic Industry Association): MII-affiliated; industry statistics and national standards. CCSA (China Communications Standards Association): Aligns China standards with ITU-T/3GPP/IEEE. ODCC (Open Data Center Committee): ByteDance, Alibaba, Baidu-founded; open standards for data center equipment; ODCC white papers on transceiver specs are becoming de facto domestic procurement requirements.

Chapter 4　China Market Size and Operating Conditions

4.1　China Optical Transceiver Market Size

China is the world's largest transceiver producer and the world's second-largest consumer (after the US).

Production scope: Chinese companies produce approximately 50%+ of global transceiver volume (by count). This is the broadest definition of "the Chinese optical transceiver industry."

Domestic consumption scope: Actual Chinese domestic consumption ~RMB 400–600 billion (approx. USD 55–80 billion) in FY2025, ~30%–35% of global. Notably: Innolight's overseas revenue was 86.81%; Eoptolink's 78.7%—the Chinese transceiver industry is highly export-driven; domestic consumption is a minority of production.

4.1.1　China Domestic Transceiver Consumption Mix (2025 Est.)

Application	Share of Domestic Consumption	Key Customers	Key Products
5G fronthaul/midhaul/backhaul	~35%–40%	Three major carriers	25G SFP28, 100G QSFP28
FTTH fiber access	~20%–25%	Three major carriers	GPON 1G / XGS-PON 10G
Domestic DC (AI/cloud)	~25%–30%	ByteDance, Tencent, Alibaba, Huawei Cloud	400G/800G QSFP-DD/OSFP
Enterprise network / other	~10%–15%	Government/finance/manufacturing	10G/25G/100G

4.1.2　Domestic Datacom Demand Rising Rapidly

ByteDance's GPU procurement ranked global top-3 in 2024–2025; Alibaba Cloud, Tencent Cloud, Baidu Cloud, and Huawei Cloud all significantly increased AI infrastructure buildout, pushing domestic 800G module deployments higher.

4.2　China Industry Overall Operating Data (FY2025)

2024 marked the fastest growth period ever for China's optical transceiver industry:

Company	Code	2024 Revenue (CNY bn)	YoY	Net Profit (CNY bn)	Net Profit YoY
Innolight	300308	23.862	+60.25%	5.171	+108.78%
Eoptolink	300502	8.647	+187.29%	2.838	+235.89%
HGtech Photonics	002281	8.272	+44.20%	0.661	+43.10%
Accelink Technology	300394	3.252	+58.79%	1.344	+50.15%
Cambridge Industries	603083	3.652	+18.31%	0.167	+75.42%
Broadex Technologies	300548	~1.5–2.0	High growth	—	+75%+

4.3　Domestic Substitution Rate by Speed Tier

Component	Speed Application	Domestic Rate (2025 Est.)	Key Bottleneck
25G DFB laser chip	25G SFP28	~70%–80%	Already broken through
100G DFB laser chip	100G SFP28/CFP2	~40%–50%	In progress
100G EML laser chip	100G/400G DR4	~30%–40%	Changelight / YJSemi in production
200G+ EML laser chip	800G/1.6T	<20%	Core chokepoint
VCSEL laser chip	Short-reach DC	~50%+	Good progress
CW laser (100 mW+)	SiPh 800G/1.6T	~20%–30%	YJSemi breaking through
DSP chip (400G/800G PAM4)	Datacom modules	<5%	Marvell/Broadcom monopoly
PIN/APD photodetector	All speeds	~40%–60%	Good progress
InP substrate	All lasers	<15%	Critical strategic gap
AWG wavelength-division mux	DWDM/100G+	~70%+	Sivers Photonics breakthrough

4.4　Industry Concentration

CR2 (Innolight + Eoptolink) 2024: Revenue ~RMB 32.5 billion, ~72% of listed optical-transceiver companies' total. CR5 (+ HGtech Photonics + Accelink + Cambridge): ~85%+. Drivers: scale economies, customer qualification barriers (12–18 months), R&D velocity, capital barriers for 800G/1.6T automation lines.

4.5　Telecom-Side Demand Deep Analysis

5G fronthaul: 3+ million 5G base stations by end-2024; each requires 3–6 pairs of 25G SFP28 fronthaul modules → steady annual replacement demand of ~1.5–2.5 million units/year.

FTTH: Domestic build largely saturated (> 570 million fixed broadband subscribers); incremental demand from overseas markets (Southeast Asia, Africa, Latin America).

DWDM coherent: Carrier backbone DWDM 400G ZR/ZR+ coherent modules—HGtech Photonics is China's main participant; Coherent and Lumentum retain leads for ultra-long-haul (>2,000 km) and 600G ZR+.

4.6　2025 H1 Industry Update

Innolight Q1 2025: revenue CNY 6.674 billion (+37.82%), net profit CNY 1.583 billion (+56.83%)—growth decelerating from 2024 double-digit pace on high base, but absolute scale at historical high. Eoptolink H1 2025 silicon-photonics and LPO product customer validations progressing well. US tariff impact (145% rate on some categories) tested in H1 2025: hyperscale customers appear to absorb costs rather than switch suppliers; Chinese export volumes showed no material decline.

4.7　Tariff Impact Analysis

US tariff hikes in 2025 on optical transceivers (HS 8517.62 class) create complex supply-chain dynamics. Hyperscale customers (Meta, Google, Microsoft) used purchasing scale and DDP contract structures to partially share tariff costs; import tariff exemption applications filed with USTR. Short-term "pull-forward" buying (pre-tariff stockpiling) inflated Q1 2025 shipments; brief normalization lag in Q2–Q3. No evidence of structural customer defection to Western suppliers—which would require years of qualification cycles and non-existent alternative capacity.

Chinese vendor responses: exploring Vietnam/Malaysia final-assembly options; deepening technical lock-in through joint 1.6T LPO co-development with customers; expanding non-US markets (Europe, Middle East, Southeast Asia).

4.8　China Optical Transceiver IP Landscape

Lumentum (via Oclaro) and Coherent hold key EML laser fundamental patents. Marvell and Broadcom hold extensive PAM4 DSP algorithm patents. LPO solutions partially circumvent DSP patents but generate their own IP thickets. Chinese companies have significantly increased patent filings in packaging, high-speed board-level interconnects, and automated testing—but face gaps in core photonic device (laser chip, modulator principle) IP vs. Western leaders. OCP (Open Compute Project) has lowered IP barriers by promoting open design specs.

Chapter 5　Supply-Chain Deep Dive

5.1　Supply-Chain Structure and Value Distribution

Value distribution favors upstream: InP/GaAs substrates (50%–70% gross margin), EML/DFB laser chips (40%–60%), DSP/driver chips (45%–65%), optical components/OSA (35%–50%), module assembly—leading vendors (30%–40%), smaller vendors (15%–25%), optical fiber cable (20%–30%).

The value gradient explains why even Innolight (35% gross margin) trails Marvell (50%+ DSP gross margin)—largest scale does not equal highest margin. This is the economic incentive behind Chinese module companies' upstream vertical integration moves (self-developed EML chips, DSP programs).

5.2　Upstream I: Substrate Materials

5.2.1　InP—The Strategic Chokepoint of Optical Communications

InP's direct bandgap enables efficient laser emission and modulation in the 1310 nm and 1550 nm low-loss windows—something silicon cannot do directly. As AI drives demand for 800G/1.6T EML-based modules, InP substrate has become the "strategic chokepoint" of the optical communications supply chain.

Global InP substrate market (2024E): ~USD 0.7–1.0 billion (including epitaxial wafer), growing ~20%–25% per year. Coherent Corp (US, ~40%–50% share), Sumitomo Electric (Japan, #2), Wafer Technology (UK, now part of Coherent), AXT Inc. (US, with production at Beijing Tongmei Xtal Technology).

China domestic substitution: Below 15%; most domestic InP substrates are sub-4-inch, with defect density and uniformity gaps vs. foreign products. A five-year+ project to achieve meaningful substitution.

5.2.2　GaAs and Silicon Substrates

GaAs is used for VCSEL and some DFB lasers; domestic substitution ~40%–50%. Silicon (for SiPh) is CMOS-compatible and manufactured at foundries like TSMC—but still requires an external InP-based CW laser source.

5.3　Upstream II: Laser Chips

5.3.1　DFB Lasers

YJSemi (688498) has achieved large-scale commercial production of 25G DFB chips, breaking Lumentum/Broadcom monopoly in 5G fronthaul supply; 100G DFB domestic rate rising to ~40%–50%.

5.3.2　EML Lasers

EML integrates DFB laser and electroabsorption modulator on a single InP die—the highest-difficulty photonic chip in the transceiver supply chain. Requires MBE/MOCVD InP epitaxy with nanometer-layer precision, sub-0.1 µm grating lithography, and complex modulator–laser interface optimization.

Global leaders: Lumentum (Oclaro IP), Coherent, Sumitomo, Fujitsu. Domestic progress: Changelight (688048) has 100G EML in mass production; YJSemi has 100G EML in customer validation and 200G EML developed.

5.3.3　VCSEL Lasers

For multimode short-reach datacenter interconnects. Lower technical barrier; domestic substitution ~50%–60%.

5.3.4　CW Lasers

Essential light source for silicon-photonic solutions. YJSemi's 70 mW CW chip is in mass delivery; 100 mW CW chip has completed customer validation; 400G/800G SiPh module CW chips exceeded 1 million units shipped in 2024—a meaningful domestic breakthrough.

5.4　Upstream III: DSP and Driver Chips

Marvell (PAM4 DSP Alaska/Spica series) + Broadcom (OPB series) dominate global 400G/800G DSP (~70%–80% combined). LPO eliminates DSP entirely—the key countermeasure to supply-chain dependency on Marvell and Broadcom, applicable to short-reach (≤500 m) AI cluster internal interconnects.

5.5　Midstream: Optical Component (OSA) Assembly

Active alignment (6-axis sub-micron precision coupling of EML output to 9 µm SMF core, laser spot welding fixation) is the most critical precision manufacturing step—accumulated device parameter databases and equipment tuning experience are non-replicable competitive moats.

Accelink Technology (300394): 2024 revenue CNY 3.252 billion (+67.74%), net profit CNY 1.344 billion (+84.07%), net margin ~41%—confirming the premium value of the component tier.

5.6　Midstream: Optical Fiber Cable

Long-distance fiber consumption grows with AI network scale. Yangtze Optical Fibre and Cable (61368) is the global top-3 producer; Hengtong Photonics, ZTT, and Tongding Interconnect are significant players. China's fiber cable capacity accounts for 60%+ of global production—another sub-segment with decisive Chinese dominance.

5.7　Downstream: AI Servers and Network Equipment

An NVIDIA DGX H100 rack requires several 400G/800G transceivers; a 10,000-card H200 cluster consumes 5,000–100,000 modules. Broadcom Tomahawk 5 (51.2 T) → 800G; Tomahawk 6 (2025) → 1.6T/3.2T. NVIDIA InfiniBand (QM9700, QM9300) and Spectrum-X platforms are key demand drivers. China's domestic switch vendors (Huawei CloudEngine, H3C, ZTE) are important sources of domestic 800G transceiver demand.

5.8　Precision Manufacturing Process—The Hidden Moat

Auto-active alignment, TO-CAN packaging and laser welding, 800G out-of-box testing (optical power ±0.5 dBm, ER >7 dB @ 100 GBaud, BER <10⁻⁴ PRBS-31, 0°C–70°C temperature sweep): a complete 800G test run takes 3–5 minutes; test equipment (Keysight N7015A) costs CNY millions each; accumulated process know-how is difficult to replicate quickly.

5.9　Cost Structure Analysis

For an 800G DR8 module: EML laser chips (30%–40%), DSP chips (20%–25%), driver/TIA chips (5%–8%), photodetectors (5%–10%), housing/PCB (8%–12%), test/yield losses (5%–10%), other (~5%–8%). EML + DSP together account for ~50%–65% of total BOM cost—both from foreign sources. Domestic EML breakthrough will sharply improve module manufacturers' gross margins.

5.10　New Product Lines: AOC, DAC, and Optical Engines

800G/1.6T AOC adoption is growing in AI cluster high-density deployments. Optical engines (the CPO core component) are being developed by ZTE Micro, Huawei Photonics, and YJSemi (CW laser). Optical engine manufacture will require deep technical partnerships with Broadcom/NVIDIA—simultaneously an opportunity (entering higher-value nodes) and a challenge (requiring more complex ecosystem relationships).

5.11　Environmental Reliability and Lifecycle Management

Design lifetime ≥7–10 years (MTTF >10⁷ hours for EML lasers). HALT/HASS accelerated aging tests, hot-swap FRU replacement capability (no service interruption). "Consumable" attribute drives recurring procurement demand beyond initial infrastructure builds.

5.12　Energy Efficiency and Green Computing Requirements

800G transceiver power: 14–20 W (~17–25 pJ/bit); 1.6T target: 12–18 pJ/bit; CPO target: <10 pJ/bit. NDRC PUE ≤1.3 mandate for new large data centers drives LPO/CPO deployment. Green MSA initiatives push low-load power saving modes. Carbon footprint (PCF) reporting will become a procurement consideration within 3–5 years.

Chapter 6　Key Company Deep Dive

6.1　Innolight (300308)—Global Optical Transceiver #1

Headquartered in Jinan (Shandong); primary manufacturing base in Suzhou. Formed by the 2018 merger of Zhongji Equipment and Innolight Technology (founded 2003 by Silicon-Valley-returnee engineering team).

2024 Financials: Revenue CNY 38.24 billion (+60.25%), net profit CNY 10.797 billion (+108.78%), net margin ~28.2%. Overseas revenue CNY 20.715 billion (86.81%); R&D expenditure CNY 1.333 billion (+64.64%, 1,453 R&D staff). Q1 2025: revenue CNY 6.674 billion (+37.82%), net profit CNY 1.583 billion (+56.83%). Ten-year growth: ~861×.

Product leadership: 400G global share ~50%; 800G global share ~30%+; 1.6T—world's first mass-production batch shipped. Full speed coverage 10G–1.6T.

Competitive advantages: Hyperscale customer lock-in (multi-year framework contracts + new speed tier priority supply); first-mover technology record (first 800G mass production, first 1.6T shipment); Suzhou manufacturing moat (automated active alignment, high-yield process databases).

Risks: High overseas revenue concentration (87%), sensitivity to US policy; 800G competitive landscape intensifying; long-term DSP dependency on Marvell/Broadcom.

6.2　Eoptolink (300502)—Highest-Growth Challenger

Headquartered in Chengdu, Sichuan; manufacturing based in Chengdu.

2024 Financials: Revenue CNY 24.842 billion (+187.29%), net profit CNY 9.532 billion (+235.89%), net margin ~38.4% (higher than Innolight). Overseas revenue CNY 6.805 billion (78.7%). Total assets CNY 12.267 billion (+90.47%); net assets CNY 8.328 billion (+52.36%). R&D expenditure CNY 403.17 million (4.66% of revenue).

Technology differentiation: Eoptolink adopted silicon-photonics and TFLN (thin-film lithium niobate) earlier than most domestic rivals, releasing 400G/800G/1.6T module families based on SiPh and TFLN platforms. Also active in LPO and AEC (active electrical cable) for ultra-short-reach AI cluster interconnects.

Market/customers: Global top-5 hyperscale customers (partially overlapping with Innolight); ByteDance is an important domestic customer. New customer wins through hyperscale multi-sourcing strategy contributed significantly to 2024 growth.

6.3　HGtech Photonics (002281)—State-Owned Wuhan Optics Valley Leader

Wuhan-based; spun off from Wuhan Post & Telecom Research Institute; state-owned shareholder (Fiberhome Technology Group).

2024 Financials: Revenue CNY 11.929 billion (+44.20%), net profit CNY 661 million (+43.10%), net margin ~7.99%. R&D expenditure CNY 713 million (+27.41%).

Product scope: Full spectrum of optical components (AWG, EDFA, TO-CAN, PLC splitter) + modules (datacom + telecom)—unique "vertical integration" among Chinese transceiver companies. Investee Sivers Photonics (AWG chips) has broken the foreign monopoly.

Datacom ramp: 400G monthly capacity 40–45 million units by end-2024 (80%+ utilization), expanding to 70 million/month. 1.6T in sample testing, targeting 2025 mass production.

Huawei relationship: HGtech Photonics is a key supplier of optical components and modules for Huawei's data center and 5G equipment—distinguishing its domestic/Huawei-ecosystem customer mix from Innolight's and Eoptolink's US-hyperscale focus.

Medium-term logic: Domestic AI datacom boom (ByteDance/Tencent/Alibaba/Huawei Cloud AI infra) will benefit HGtech most among the top three; vertical integration provides better price-war resilience vs. pure module assemblers.

6.4　HGtech Zhongyuan (under HGtech Science & Technology, 000988)

HGtech Science & Technology (A-share 000988, spun off from HUST) subsidiary HGtech Zhongyuan (Wuhan HGtech Zhongyuan Photonic Technology) is a key global datacom transceiver supplier.

400G capacity 40–45 million units/month (80%+ utilization), expanding to 70 million/month. Full 200G/400G/800G delivery capability. HGtech Science & Technology group revenue: ~CNY 90 billion in Q1–Q3 2024 (+23.42%).

6.5　Cambridge Industries Group (603083)—Triple-Business Layout

Shanghai-headquartered; broadband access (ONT/CPE), wireless access (Wi-Fi router), and high-speed transceiver modules.

2024 Financials: Revenue CNY 3.652 billion (+18.31%), net profit CNY 167 million (+75.42%). High-speed transceiver (mainly 100G/400G datacom) is the primary growth engine; Wuhan factory (Shanghai Accelink Technology Wuhan Co.) sits within the Wuhan Optics Valley ecosystem.

6.6　Accelink Technology (300394)—High-Margin Optical Component Leader

Suzhou-headquartered; passive optical components (connectors, splitters, isolators), active sub-assemblies.

2024 Financials: Revenue CNY 5.163 billion (+58.79%), net profit CNY 2.017 billion (+50.15%), net margin ~41%—one of the highest in the optical communications sector, confirming the premium of the component tier.

6.7　YJSemi (688498)—Optical Chip Domestic Substitution Pioneer

YJSemi (Shaanxi YJSemi Semiconductor Technology), STAR Market 688498; the only Chinese company to achieve commercial mass production of DFB/EML/CW laser chips.

Key milestones: 25G DFB—mass production; 100G PAM4 EML—customer validation completed (2024); 200G PAM4 EML—product developed, in market; CW laser (70 mW)—batch delivery; CW (100 mW)—customer validation completed; 400G/800G SiPh module CW chips—>1 million units shipped in 2024 (breaking foreign monopoly).

Strategic significance: By supplying CW lasers, YJSemi provides the national-domestic foundation for 800G/1.6T silicon-photonics modules; EML commercialization represents China's step from "buying chips" to "making chips."

6.8　Changelight (688048)—EML Mass-Production Second Pillar

Suzhou Changelight Photonics Technology, STAR Market 688048; pedigree from Changchun Institute of Optics and Mechanics (CAS).

Key milestones: 100G EML—mass production shipped (domestic largest volume); 200G EML—samples sent (expected 2025 mass production); 100G VCSEL and CWDM4 CW laser—pending mass production.

6.9　Company Comparison Summary

Company	Code	Positioning	2024 Revenue	Gross Margin	Core Competence
Innolight	300308	Global module #1	CNY 23.862 bn	~35%	Customer moat + scale mfg + iteration speed
Eoptolink	300502	Fastest-growing challenger	CNY 8.647 bn	~35%+	SiPh + LPO multi-track technology
HGtech Photonics	002281	State-owned telecom leader	CNY 8.272 bn	~28%	Vertical integration + domestic telecom channels
HGtech Zhongyuan	(under 000988)	Datacom capacity expansion	—	—	Wuhan Optics Valley ecosystem + HGtech group
Accelink Technology	300394	High-margin component leader	CNY 3.252 bn	~55%	Precision component moat
Cambridge Industries	603083	Triple-track diversified	CNY 3.652 bn	~20%	Broadband + wireless + module balance
YJSemi	688498	Optical chip pioneer	—	~45%+	DFB/EML/CW full-series production
Changelight	688048	EML mass-production pillar	—	~50%	CAS InP epitaxy technology

6.10　Mingpu Optoelectronics (002902) and Taitwah Communications (300570)

Mingpu Optoelectronics (Dongguan, A-share 002902): Integrated magnetics and optical transceiver manufacturer; mainly 10G–100G with datacom 400G expansion. Competitive positioning on price/performance.

Taitwah Communications (300570): Suzhou-based; optical fiber adapters, connectors (LC/SC/MTP), WDM filters; specializes in fiber connectivity accessories. Benefiting from high-density cable infrastructure demand in AI datacenters.

6.11　Emerging Companies to Watch

Wuhan Minxin Semiconductor: DFB/VCSEL laser and PIN/APD detector manufacturer; Wuhan Optics Valley ecosystem; dual-track optical communications + LiDAR growth engine.

Sivers Photonics (investee of HGtech Photonics): AWG chip specialist; broke foreign monopoly in DWDM wavelength-division components.

ZTE Micro Electronics: SiPh modulator chips (currently internal supply to ZTE), important window into China's SiPh commercialization progress.

Photonic integration startups: NioKem, Guangzi Arithmetic, Xizhi Technology, etc.—active STAR/VC-backed frontier innovation.

6.12　Optical Transceivers and AI Chip Ecosystem

NVIDIA: indirect demand driver via GPU server shipments + CPO technology direction setter (InfiniBand OSFP standard). Huawei: both a major optical transceiver customer and a competitor (via its own optical components and modules division); Huawei's SiPh chip R&D could reshape the domestic photonic chip ecosystem. Domestic AI chip ecosystem (Huawei Ascend, Cambricon, Biren) creates a "domestic AI chip → domestic compute cluster → domestic transceivers" positive feedback loop.

6.13　Innolight Growth Story: From Motor Manufacturer to Global #1

2014: revenue ~CNY 27 million (standalone Innolight Technology). 2018: merged with Zhongji Equipment (A-share listing access). 2024: revenue CNY 38.24 billion—10-year growth of ~861×. This trajectory combined AI era demand explosion with a returnee Silicon Valley engineering team's prescient early layout in high-speed transceivers.

6.14　Eoptolink: Western China "Tech Manufacturing" Rise

Eoptolink is a homegrown Chengdu enterprise built by local engineers without returnee founders. Its early (2021–2022) SiPh/TFLN technology bets—earlier than Innolight's public disclosures—explain the 33% net margin exceeding Innolight's 22% in 2024. Chengdu government incentives (land, tax, talent subsidies) amplified the scale expansion pace.

Chapter 7　China Optical Communications Industrial Cluster Distribution

7.1　Cluster Structure Overview

China's optical communications industry (transceivers, optical components, fiber cable, optical chips) is highly geographically concentrated: Wuhan, Suzhou, Shenzhen/Dongguan, Chengdu, and Shanghai/Beijing form the five key zones, with Wuhan Optics Valley dominant. Geographic concentration reflects the co-location of R&D, precision manufacturing ecosystems, and talent pools—not easily replicated by new entrants.

The optical transceiver industry is highly export-oriented (leading manufacturers 80%–90% overseas revenue), meaning these clusters serve global AI compute infrastructure buildout rather than primarily local demand. The platform's factory coverage enables supply-chain traceability and capacity verification across all major clusters.

7.2　Wuhan Optics Valley—China's Largest Optical Communications Base

Wuhan East Lake High-Tech Development Zone (Wuhan Optics Valley) is both China's absolute optical communications core and globally one of the most important optical manufacturing bases. The origin: in 1976, Academician Zhao Zisen of the Wuhan Post & Telecom Research Institute drew China's first practical optical fiber—making him "China's Father of Optical Fiber." Over 40+ years, Wuhan has continually incubated optical fiber cable, optical component, and optical communications equipment champions.

Scale (2024): Hubei Province's optoelectronic information industry broke through the CNY 1 trillion mark in 2024—the province's first pillar industry to exceed CNY 1 trillion. Wuhan's optical communications products account for ~50% of the national market and ~25% of the global market (by output value).

Core enterprise cluster: HGtech Photonics (002281), Fiberhome Communications (633876), Yangtze Optical Fibre (61368), HGtech Zhongyuan, Wuhan Minxin Semiconductor, Sivers Photonics (investee of HGtech), Changelight (688048, with Wuhan technical roots).

Policy support: Full-spectrum: land, R&D subsidies, talent programs; 2024 "China Optics Valley Optoelectronic Information Industry Innovation Development Conference" further strengthened ecosystem cohesion.

Global ranking: Largest optical fiber cable production base globally; largest optoelectronic device production base in China; largest mid-small display panel R&D/manufacturing base in China.

7.3　Suzhou—Optical Transceiver Precision Manufacturing Center

Suzhou Industrial Park is China's highest-density precision electronics manufacturing zone. Innolight's primary manufacturing base operates here on the strength of a mature SMT placement ecosystem (JUKI/ASM maintenance centers), precision mechanical sub-suppliers (module housings, spring contacts, precision stamped QSFP-DD latch parts), and sub-micron active-alignment process capabilities.

Accelink Technology (300394) is headquartered in Suzhou, leveraging its precision mechanical manufacturing expertise at micrometer-to-sub-micrometer tolerances.

7.4　Shenzhen/Dongguan/Guangdong—Commercial Ecosystem

Proximity to Huawei (world's largest optical communications equipment vendor), Shenzhen Stock Exchange, and overseas procurement offices drives commercial ecosystem concentration. Mingpu Optoelectronics (Dongguan), Broadex Technologies (Shenzhen branch), and many SME transceiver manufacturers serve 10G–100G and access-side segments.

7.5　Chengdu/Sichuan—Emerging Western Hub

Eoptolink's headquarters and core manufacturing base in Chengdu, supported by UESTC optical engineering talent pool and Western China manufacturing cost advantage. Eoptolink's CNY 8.647 billion revenue (+179%) in 2024 validates the full competitiveness of Western China optical transceiver manufacturing. Sichuan Province has designated optical communications as a priority "new information technology industry," providing additional policy support.

7.6　Shanghai/Beijing—R&D Centers

Shanghai: Cambridge Industries Group headquarters; Fudan University / Tongji University photonics research centers; semiconductor device equipment companies. Beijing: YJSemi R&D center; CAS Institute of Physics; Peking University photonics lab; Tsinghua TFLN research.

7.7　Industrial Cluster Comparison

Cluster	Core Strength	Representative Companies	Supply-Chain Position
Wuhan Optics Valley	Deepest ecosystem, longest history, 25% global output	HGtech Photonics, Fiberhome, Yangtze Optical, HGtech Zhongyuan, Wuhan Minxin	Telecom components+modules+fiber+chips
Suzhou	Precision manufacturing, automation, largest capacity	Innolight (mfg base), Accelink Technology	High-speed datacom modules + high-margin components
Shenzhen/Dongguan	Commercial ecosystem, Huawei proximity, trade convenience	Mingpu Optoelectronics, Broadex, many SMEs	Mid-low speed modules + distribution/trade
Chengdu/Sichuan	Western cost, ample talent, emerging rise	Eoptolink, Huatuo Optical	High-speed datacom module fast ramp
Shanghai/Beijing	R&D centers, university research, frontier tech	Cambridge (Shanghai), YJSemi (Xi'an+Beijing)	R&D output + chip breakthrough

China's optical communications cluster landscape is "Wuhan + Suzhou + Chengdu" as the three pillars. Wuhan defends the "China's Optics Valley" historical moat; Suzhou produced global #1 Innolight via precision manufacturing; Chengdu leaped from zero to CNY 10 billion via Eoptolink, becoming the western manufacturing sector's most dazzling new pole.

7.8　Upstream and Downstream Ecosystem and Support Infrastructure

Wuhan: 3,000+ optical communications-related companies (including many specialty SMEs); MOCVD equipment local maintenance; precision optics (lens grinding, coating); fiber preform drawing (Yangtze Optical); system integration (Fiberhome, HGtech). Suzhou: JUKI/ASM placement machine maintenance centers; precision laser welding equipment (YAG weld for TO-CAN fixation); AOI automated optical inspection; precision casting (module housings); spring terminal manufacturers. Chengdu: Intel Chengdu assembly/test factory + BOE have built precision manufacturing baseline; UESTC talent pipeline is deep.

7.9　Talent Structure and Engineer Market

Wuhan: HUST (Huazhong University of Science and Technology) trains hundreds of optical-fiber/device M.S./Ph.D. candidates annually; Wuhan Post & Telecom Research Institute is the career origin for thousands of Wuhan Optics Valley engineers.

Suzhou: Returnee Silicon Valley engineering talent (Innolight founding team + subsequent hires); mid-level salary structure attracting both premium engineers and qualified production workers at competitive cost vs. Beijing/Shanghai/Shenzhen.

Chengdu: UESTC provides nationally top-tier microelectronics, communications engineering, and optoelectronics graduates; Eoptolink has strong hometown recruitment appeal.

7.10　International Comparison: Wuhan vs. Japan / US / Europe

vs. Japan (Sumitomo, Fujitsu): Japan has deep technical heritage but faces structural hollowing (aging, high labor cost, engineer shortages); Japan focuses on high-end telecom-side, coherent modules, and specialized materials—no serious participation in bulk datacom module manufacturing.

vs. US (Coherent/Lumentum locations): US strength in technical innovation (EML, DSP, CPO) and upstream materials; but Coherent's fragmented multi-country manufacturing (Pennsylvania, China, Europe) cannot match Wuhan's ecosystem cohesion efficiency.

vs. Taiwan: TSMC's SiPh platform (CPO era key chip) is Taiwan's strongest optical communications card; but this is semiconductor foundry, not optical transceiver module manufacturing.

Overall: Wuhan + Suzhou + Chengdu constitute a globally irreplaceable optical transceiver manufacturing cluster, built over 30+ years of accumulation. The co-location synergies of supply, talent, and knowledge spillovers will continue to raise the entry barrier for challengers.

7.11　Digital Transformation Impact

Wuhan Optics Valley industrial internet platforms; Innolight Suzhou factory AI vision guidance for active-alignment robots; key equipment domestic substitution (precision placement equipment from domestic providers like Jiepu, Dier Laser). Long-Yangtze supply-chain digital visibility platforms linking Innolight and Accelink to hundreds of sub-suppliers for order visibility, inventory sharing, and quality data integration.

Chapter 8　Sub-Segment Specialty Research

8.1　800G—The Current Most Important Growth Driver

8.1.1　Technical Background

800G uses 8 channels × 100 GBaud PAM4 in QSFP-DD or OSFP form factors. The transmit side requires 8 EML laser chips (vs. 4 for 400G), directly doubling EML demand per module—the root cause of 800G-era EML supply tightness.

8.1.2　2024 Demand Explosion Data

• Global 800G shipments (2024): ~7.5 million units (from <1 million in 2023); YoY growth ~6×+.
• 2025 forecast: >20 million units (LightCounting est.).
• Key driving customers: Meta, Google (YouTube DC expansion), Microsoft (Azure OpenAI capacity), Amazon AWS, ByteDance, Tencent Cloud.
• NVIDIA Blackwell effect: GB200 and H200 GPU server configurations adopt 800G OSFP transceivers; Blackwell's large-scale deployment triggered step-change 800G demand.

8.1.3　800G Market Competitive Landscape

Innolight: 800G global share ~30%+; first-mover and highest-volume single vendor. Eoptolink: entered global 800G top-3 by mid-2024. HGtech Zhongyuan: substantial portion of 400G capacity already converted to 800G. Coherent + Lumentum combined: ~40% of Western supplier 800G share, ~15%–20% of global total.

8.1.4　800G Price Trajectory

ASP: ~USD 300–350 (2023) → ~USD 130-180–250 (2024) → ~USD 130–180 (2025E). Falling ASP is offset by 6× demand growth—total 800G market size expanded strongly in 2024 despite price erosion.

8.2　1.6T—The Next Mass-Production Node

8.2.1　Technical Routes

Option 1 (DSP, OSFP/CPO): 16 × 100 GBaud PAM4 or 8 × 200 GBaud PAM4 with 200G EML or silicon-photonic modulator + new-generation Marvell Spica 2 DSP.

Option 2 (LPO): No DSP; linear driver directly drives laser; lower power/latency; suited to ≤500 m ultra-short-reach AI cluster internal interconnects. Innolight, Eoptolink demonstrated 1.6T LPO OSFP samples at OFC 2024.

Option 3 (CPO): TSMC CPO modules in production from 2025.

Report judgment: 1.6T technology routes will converge in 2025–2026: DSP dominant for most scenarios, LPO for ultra-short-reach, CPO for 1–2 flagship data center pilots.

8.2.2　1.6T Mass-Production Progress

• 2024: Innolight, Eoptolink—sample delivery to AI customers; HGtech Photonics—in sample testing.
• H1 2025: China-produced 1.6T modules accounted for ~45% of global shipments.
• 2026 forecast: Global 1.6T demand >10 million units.
• 2027: 1.6T may enter co-dominance with 800G, pending 3.2T/CPO commercial ramp.

8.3　CPO—Long-Term Technology Shift

CPO eliminates the PCB trace length between switch ASIC and pluggable module socket, sharply reducing signal loss and power (30%–40% savings).

Commercialization timeline: TSMC CPO module production started 2025 (sample delivery to Broadcom and NVIDIA); limited CPO deployments at 1–2 flagship data centers expected 2026; CPO penetration reaching meaningful levels (10%+ of new DC installations) not until 2027–2028.

China position: Innolight, Eoptolink have stated CPO R&D investments; ZTE Micro and Huawei photonics teams are pursuing SiPh CPO; 2025–2027 window is sufficient to complete CPO technical reserves while fully shipping pluggable modules.

8.4　Coherent Modules—Telecom Long-Reach Technical Moat

Coherent optical modules use phase and polarization of light (QAM modulation) for spectral efficiency, enabling thousands-of-km regeneration-free transmission. Coherent, Lumentum, HGtech Photonics (400G ZR) are global main players. China lags behind Western vendors in super-long-reach (>2,000 km) and 600G ZR+ coherent—the remaining segment where China has not achieved market dominance.

8.5　Datacom vs. Telecom: Market Driver Comparison

Dimension	Datacom	Telecom
2024 market share	50%+	35%
2024 growth rate	+80%–100%+	~+10%–20%
Primary driver	AI compute cluster build	5G deep coverage, FTTH infill
Speed frontier	800G/1.6T (rapid iteration)	100G–400G (relatively stable)
Price trend	ASP falling fast but volume offsets	Relatively stable

8.6　AOC and DAC

AOC (integrated cable + module ends) competitive advantages growing in high-density AI cluster deployments due to light weight and superior cable density. DAC (copper passive) dominant for ≤7 m extreme short-reach; application scope relatively stable.

8.7　AI Compute Cluster Topology and Transceiver Demand

Spine-Leaf Ethernet: Leaf TOR → 16 × 400G uplinks to Spine; 64 Spine switches × 128 × 800G ports per switch. A 10,000-card H200 cluster requires ~16,000 × 400G + 8,192 × 800G = ~25,000 modules (+ AOC + storage network + OOB management), total 30,000–50,000 modules per 10K-GPU cluster, single-order value ~USD 50–100 million.

InfiniBand NDR 800G (OSFP): Used by Meta AI, Microsoft Azure AI, OpenAI training clusters; OSFP NVIDIA adoption is why OSFP became 800G critical. Innolight strong in IB ecosystem.

FC-SAN storage: 32G/64G FC modules in large AI cluster storage networks; steady demand for HGtech Photonics and similar vendors.

8.8　Standards Organizations

MSA families: QSFP-DD MSA, OSFP MSA, CMIS 4.0/5.0, CPO MSA. IEEE 802.3 working groups: 802.3bs (400G), 802.3ck (100 GBaud), 802.3df (1.6T—in progress). OIF: 400ZR/ZR+ coherent, CPO electrical interface specs. ITU-T: DWDM wavelength grid standards.

8.9　Secondary Market and Refurbishment

Active 100G/400G refurbished transceiver market (20%–40% of new price); US Amazon/eBay platforms active; Shenzhen distributors also trade refurbished units. ~5%–10% of global shipments by volume. Chinese-branded 400G (especially Innolight) increasingly common in refurbished market, further strengthening brand penetration in global data center operations ecosystems.

8.10　International Spec Differences: Commercial vs. Industrial Grade

Commercial grade (0°C–70°C) for controlled-temperature hyperscale DCs; industrial grade (−40°C to +85°C) for outdoor 5G base stations—higher manufacturing cost, lower margin. 800G QSFP-DD EAR99 classification allows unrestricted US export currently; reclassification remains a latent risk.

8.11　LiDAR Technology Crossover

YJSemi, Wuhan Minxin semiconductor laser chips serve both optical communications and LiDAR (905/1550 nm). LiDAR market ~USD 2–3 billion globally (2024E), CAGR ~20%–30%—a meaningful secondary growth engine for laser chip manufacturers. Different parameter requirements (ultra-short pulse, AEC-Q102 automotive-grade) prevent simple cross-use; require dedicated product lines.

8.12　New Markets Beyond the US

Europe: AWS, Microsoft Azure, Google Europe data center expansions; NIS2 directive compliance requirement manageable. Middle East: NEOM/STC (Saudi Arabia), Microsoft + G42/AWS Middle East (UAE)—fastest-growing new data center market 2025–2030; Chinese vendors have some ecosystem advantage via Huawei. Southeast Asia: ByteDance, Alibaba (SEA operations), Grab, Sea Group—highly supply-chain-integrated with China; highest-potential non-US emerging market for Chinese transceiver vendors.

Chapter 9　Technology Evolution Roadmap

9.1　Technical Logic of Speed Upgrades

Modulation evolution: NRZ (2-level, 1 bit/Baud) → PAM4 (4-level, 2 bits/Baud) → coherent QAM (4–6 bits/symbol) → 200 GBaud PAM4 (most advanced current commercial datacom). Speed increases require wider laser modulation bandwidth (3 dB BW > 0.7× baud rate), faster DSP processing, and more demanding PCB signal integrity.

Channel count vs. baud rate tradeoff: 800G = 8 × 100 GBaud PAM4; 1.6T can be 8 × 200 GBaud (fewer channels, higher baud rate) or 16 × 100 GBaud (more channels). Former imposes tighter EML requirements; latter requires higher packaging density.

9.2　800G Technology Paths

800G SR8 (multimode VCSEL ≤100 m), DR8 (SMF EML ≤500 m, campus mainstream), FR8 (EML+WDM 2 km), LR8 (EML 10 km); QSFP-DD and OSFP both standardized—OSFP preferred for NVIDIA InfiniBand's thermal needs.

9.3　1.6T Technology Route Divergence

200 GBaud EML route: Highest modulation bandwidth demand; Lumentum and Coherent can produce; Changelight and YJSemi 200G EML in sampling/validation.

Silicon-photonic 1.6T route: SiPh modulator + CW laser (8 × 200 GBaud); avoids InP EML dependency; Eoptolink's TFLN platform is the leading domestic effort.

1.6T LPO route: No DSP; OFC 2024 multi-vendor 1.6T LPO demonstrations; targeted for ≤500 m AI cluster internal interconnects.

Report judgment: 2025–2026 convergence toward DSP-dominant + LPO supplement + CPO pilots.

9.4　Silicon Photonics—Platform Technology Transformation

SiPh replaces InP-based discrete lasers with CMOS-foundry-compatible optical modulators, detectors, waveguides—bringing optical component manufacturing into standard semiconductor fabrication. CW lasers (from InP, e.g., YJSemi) still required externally.

TSMC: Mature SiPh PDK; 3D photonic engine for CPO is the highest-profile commercial product. Intel: 15+ years in SiPh; OIO optical I/O is its commercial vehicle. China: ZTE Micro (SiPh modulator chips, internal supply); YJSemi (CW laser, already in production); Eoptolink (TFLN-based 1.6T); CAS spin-off photonic integration startups.

Penetration timeline: 400G/800G: ~15%–25% (2024–2026); 1.6T: ~30%–40% (2027–2029); CPO era: >50%.

9.5　CPO Technology Roadmap in Detail

TSMC 3D Photonic Engine roadmap (2025): 1.6 Tbps engine (for pluggable 1.6T, pre-full-CPO) → 6.4 T engine co-packaged with switch ASIC (2026) → 12.8 T+ (2027+). Power saving: ~1–3 W per CPO module vs. pluggable; USD multi-million annual electricity savings for 100K-GPU cluster. CPO TCO (total cost of ownership) expected to fall below pluggable by 2028–2030.

9.6　TFLN—Next-Generation Modulator Material

Thin-film LiNbO₃ offers >100 GHz electro-optic modulation bandwidth (vs. ~50–60 GHz for Si-based modulators), low insertion loss, high extinction ratio. NANOLN, HyperLight provide commercial TFLN wafers; Fudan, Zhejiang, and Nanjing universities lead domestic TFLN research. Eoptolink has committed TFLN as its 1.6T route option.

9.7　Technology Evolution Master Timeline

Year	Mainstream Speed	Emerging Speed	Key Milestones
2022–2023	100G/400G	800G nascent	PAM4 mature, 400G DR4/LR4 at scale
2024	400G/800G	1.6T prototypes	800G +6×, 1.6T first shipments
2025	800G/1.6T	1.6T mass production	TSMC CPO engine production; LPO route selection
2026	800G/1.6T	3.2T R&D	1.6T >10M units/yr; CPO small deployments
2027	1.6T/3.2T	3.2T ramping	CPO penetration starts affecting pluggable market
2028–2030	3.2T/CPO	—	SiPh penetration >50%; InP EML domestic breakthrough

9.8　Signal Integrity and Packaging Technology Evolution

800G PCB traces (100 GBaud): 50 Ω differential impedance, back-drill vias, low-loss laminate (Dk <3.5, Df <0.005). 1.6T (200 GBaud): even tighter requirements, next-generation ultra-low-loss materials needed. Optical coupling efficiency (EML output → 9 µm SMF core via micro-lens, ±0.5 dB yield consistency via YAG spot-weld fixation) is the primary precision manufacturing challenge and moat. Thermal management: high-conductivity baseplates, micro heat pipes, optional TEC for wavelength stabilization (±0.08–0.1 nm/°C).

9.9　Key Technology Milestone Predictions 2025–2028

2025: TSMC CPO samples → Broadcom/NVIDIA; Innolight/Eoptolink 1.6T mass production (H1); Changelight 200G EML customer validation (H2); 1.6T LPO first hyperscale deployment (Q4).

2026: 1.6T second-largest datacom speed; CPO first deployment at 1–2 hyperscale DCs; YJSemi/Changelight 100G EML domestic rate >50%; Broadcom Tomahawk 6 procurement triggers 3.2T forward buying; SiPh 800G penetration ~20%.

2027: 3.2T first OFC samples; CPO penetration ~10% of new DC installations; 200G EML domestic rate >30%; TFLN enters 1.6T/3.2T commercial use.

2028: CPO penetration ~20%–30% in new hyperscale DC projects; 3.2T mass production; InP substrate domestic rate >25%.

9.10　Quantum Communications and Optical Communications Crossover

QKD (quantum key distribution) networks rely on low-loss optical fiber and precision optical splitters—products where Yangtze Optical and HGtech Photonics already participate. But single-photon light sources/detectors (core of quantum communications) differ fundamentally from classical multi-photon transceivers, remaining the domain of dedicated quantum technology companies (e.g., QuantumCTek, Qasky). The long-term opportunity for optical transceiver ecosystem companies is in quantum repeater infrastructure—but this is beyond 2030's commercial horizon.

9.11　China Silicon Photonics Ecosystem Building

Leading academic institutions (PKU, ZJU, NJU, HUST) publishing frontier SiPh results in Nature Photonics, Science, Nature Electronics. Domestic MPW SiPh tape-out platforms under construction at CAS Institute of Microelectronics—3–5 years behind TSMC maturity. Domestic TFLN wafer production in early stage. Domestic photonic EDA tools (Synopsys OptSim/RSoft are foreign-dominated)—a potential second chokepoint analogous to semiconductor EDA for IC design.

9.12　5G-Advanced and 6G New Demand

5G-A (3GPP Rel. 17/18/19): ELAA massive MIMO antennas may require 100G fronthaul (vs. current 25G eCPRI), driving new 100G SFP28 fronthaul demand for HGtech Photonics and Cambridge Industries. XGS-PON and NG-PON2 (25G) for 5G-A rural backhaul under "rural revitalization" policy. Low-latency fronthaul requirements (< 100 µs end-to-end) favor LPO/direct-detect architectures in the telecom domain. 6G standardization expected 2028–2030; THz communications and space-terrestrial integration will create new long-term photonic connectivity demands.

Chapter 10　Risk Factor Analysis

10.1　Risk Panorama

Risk Category	Event	Near-Term Probability	Impact	Primary Affected
Demand	AI capex below expectations	Medium (~20%–30%)	Extreme	All; especially Innolight/Eoptolink
Competition	Price war accelerates, ASP falls faster	High (~50%–60%)	Medium-high	SMEs exit first; leaders' margins compress
Supply chain	Overseas DSP/EML supply cutoff	Low (~5%–15%)	Extreme (if occurs)	Marvell/Coherent-dependent module makers
Technology substitution	CPO commercializes faster, pluggable cannibalized	Low (near-term) / Medium (2028+)	Medium-high (long-term)	Traditional pluggable manufacturers
Geopolitics	Escalation; overseas customers restricted	Low-medium (~10%–20%)	Extreme (if occurs)	High-overseas-revenue leaders

10.2　Risk 1: AI Capex Below Expectations

AI hyperscaler capex sustainability risk stems from: 1) AI commercial return disappointment; 2) economic recession; 3) ASIC competition vs. NVIDIA GPU reshaping demand node composition. Historical analog: 2022 H2 ad-market pullback (Meta cut capex) → 2-quarter transceiver demand softening, 100G/400G inventory build. Current AI demand is more structural (AI as strategic necessity, speed upgrade "new-old rotation" demand) but not immune to phase softening. Head-count diversification across Meta/Google/Microsoft/Amazon/ByteDance provides partial buffering.

10.3　Risk 2: Price War and ASP Overshoot

400G ASP fell ~60% from 2020 to 2024. 800G on same trajectory. Dozens of Chinese SME transceiver manufacturers actively expanding capacity—pricing pressure ongoing. For Innolight (GM ~35%) and Eoptolink (net margin 33%), margin compression is manageable via scale offset; for SMEs, accelerating exit. Industry consolidation from dozens to 10–15 major participants is expected by 2027, with price war as catalytic accelerant—ultimately favorable to heads.

10.4　Risk 3: Upstream Supply Chain Disruption

DSP (Marvell/Broadcom): ~70%–80% combined global PAM4 DSP share. Export control expansion to high-end DSP would force LPO transition or await domestic DSP (5+ year timeline). LPO advances partially hedge near-term risk in ≤500 m scenarios but cannot replace DSP for longer links. EML / InP substrate: Not yet controlled; China domestic substitution progress (source: above) provides multi-year runway before policy risk materializes. CW laser: Largely broken through domestically (YJSemi)—risk essentially neutralized.

10.5　Risk 4: CPO Displaces Pluggable Market

CPO is the single most important long-term structural risk. Baseline assumption: CPO penetration stays below 5% of new DC optical interconnects in 2026; pluggable market window extends to 2027–2028. If TSMC CPO cost curve drops faster than expected or NVIDIA Rubin architecture (Blackwell Ultra successor) natively adopts CPO, the window could compress by 2 years. Chinese leaders (Innolight, Eoptolink) have stated CPO R&D commitments; the 2025–2027 window is sufficient for parallel pluggable shipment and CPO technical preparation.

10.6　Risk 5: Geopolitics and Overseas Customer Restriction

Innolight (87% overseas revenue) and Eoptolink (79%) face tail-risk direct revenue loss if the US imposes Chinese transceiver import restrictions or bans US hyperscalers from Chinese sourcing. Current shields: Counter-cost (US hyperscaler competitiveness would be severely damaged); optical transceivers not in AI-chip core control list; Chinese manufacturers' irreplaceability (other regions lack sufficient capacity to substitute). Domestic AI infrastructure expansion (ByteDance/Tencent/Alibaba/Huawei Cloud) provides a partial buffer; currently insufficient to replace overseas at 1:4 domestic-to-overseas ratio.

10.7　Risk Prioritization

Risk	Near-Term Probability	Impact	Priority
AI capex miss	20%–30%	Extreme	P1
Price war	50%–60%	Medium-high	P2
DSP/EML cutoff	5%–15%	Extreme	P2 (tail risk)
CPO acceleration	<5% (2yr)	Medium-high	P3 (long-term monitor)
Geopolitical restriction	10%–20%	Extreme	P2 (tail risk)

10.8　Industry Cycles and Inventory Risk

2001–2002 telecom bubble collapse: 3–4 year severe downturn (JDS Uniphase). 2022 mid-year adjustment: 2–3 quarter demand softening after Meta ad market retreat; recovered rapidly with AI investment surge in 2023. Current cycle structural differentiation: Demand source is the world's richest tech companies (balance-sheet-funded capex, strategic necessity, speed upgrade creates recurring replacement demand). Short-term inventory correction risk exists if AI model commercialization speed lags compute expansion ("AI surplus" scenario)—historically 1–2 quarter duration, not a systemic collapse.

10.9　Operating Risk: Governance and Management-Level Risks

Core team stability: Innolight's tech competitive advantage depends partly on a small number of key technical leaders. Capex expansion risk: Large-scale expansion (12–18 month equipment procurement + construction cycle) may face utilization pressure if demand softens before new capacity comes online. FX hedging: 80%+ overseas revenue exposure to USD/RMB; RMB appreciation from 7.1 to 6.5 would proportionally reduce CNY-reported revenue even with stable USD revenue. Customer concentration: Top-3 customers >50% revenue for both Innolight and Eoptolink; any single hyperscale customer supply-vendor adjustment would materially impact near-term revenues.

Chapter 11　2026–2030 Market Forecasts

11.1　Baseline Forecast Assumptions

Baseline (neutral scenario): AI capex 25%–35% CAGR (2025–2027) normalizing to 18%–22% by 2027–2030; technology sequence 800G → 1.6T (mainstream 2026–2027) → 3.2T/CPO (2028–2030); export controls not expanded to transceiver finished goods or DSP chips; China CR3 global share maintained at 50%+; 100G EML domestic rate reaches 50% by 2026, 200G EML reaches domestic production by 2028.

11.2　Global Optical Transceiver Market Forecast

Year	Global Market (USD bn)	YoY Growth	Primary Driver
2024 (actual)	17–18	+52%	800G explosion, AI compute clusters
2025E	23–26	+30%–50%	800G volume + 1.6T ramp
2026E	28–33	+20%–30%	1.6T scale-up, 800G mature
2027E	33–40	+15%–25%	1.6T mainstream, 3.2T/CPO start
2028E	38–46	+10%–20%	CPO penetration rising
2029E	42–50	+10%–15%	CPO/pluggable parallel
2030E	40–50+	—	CAGR (2024–2030) ~18%–22%

LightCounting 5-year CAGR forecast: ~22%; aggressive scenario: >USD 55 billion by 2030 (CAGR ~22%–25%); conservative scenario (AI bubble or export controls): ~USD 35 billion (CAGR ~12%).

11.3　Chinese Vendor Market Share Forecast

Chinese vendor combined global share rising from ~50% (2024) to ~55%–60% (2028 baseline). CR3 projected to rise from ~55% to ~55%–60% by 2027.

Upside drivers: 1.6T head-start (45% H1 2025); scale pricing improvements; LPO/SiPh iteration potential leadership.

Resistance factors: Coherent/Lumentum high-end coherent moats; CPO-era Western semiconductor–TSMC vertical integration (potentially bypassing Chinese module makers); hyperscale multi-sourcing pressure.

11.4　Speed Structure Forecast

Speed	2024 Revenue Share	2027E	2030E
≤100G	~20%	~10%	~5%
200G/400G	~45%	~30%	~15%
800G	~25%	~35%	~20%
1.6T	~5%	~20%	~30%
3.2T/CPO	<1%	~5%	~25%+

400G enters "mature declining" status by 2027 (today's 100G). 800G peaks ~2026–2027 then cedes to 1.6T. 1.6T becomes 2027–2028 market leader. CPO/3.2T inflection ~2028.

11.5　China Optical Chip Domestic Substitution Rate Forecast

Component	2024 Rate	2026E	2028E	2030E
25G DFB	~75%	85%+	90%+	~100%
100G DFB	~45%	60%+	75%+	85%+
100G EML	~35%	50%+	65%+	75%+
200G EML	~15%	25%+	45%+	60%+
High-end 1.6T EML	<20%	<25%	30%+	50%+
DSP (400G+)	<5%	<10%	<20%	25%+?
InP substrate	<15%	<20%	25%+	35%+
CW laser (100 mW)	~25%	40%+	60%+	70%+

11.6　Key Company 5-Year Projections

Innolight (300308): Q1 2025 annualized ~CNY 26.7 billion; full-year 2025 expected CNY 28–32 billion; if 1.6T ramps on schedule, 2027 revenue could reach CNY 40 billion+. Global #1 maintained through 2026–2027; CPO-era restructuring from 2028 onward as the new variable.

Eoptolink (300502): 2025 revenue expected CNY 14–16 billion; 2027 revenue potentially CNY 20–25 billion. Global top-3 timeline: 2025–2026.

HGtech Photonics (002281): Net margin recovery to 18%–22% target as 800G/1.6T capacity ramps through 2025–2026. Domestic AI datacom boom is the key incremental opportunity.

YJSemi (688498) + Changelight (688048): Scale far smaller than module makers today, but strategic value exceeds scale—100G/200G EML domestic breakthrough is the transformational inflection for the supply chain. 2026–2028 is the key rapid scaling window.

11.7　CAGR Summary

Metric	2024–2030 CAGR (baseline)	Key Premise
Global transceiver market	15%–22%	AI capex sustained, CPO gradual penetration
Chinese vendor revenue	20%–28%	1.6T head start, modest market share gain
China optical chip (EML/CW)	50%–80% (extreme high growth)	Domestic substitution breakthrough
CPO market (from 2027)	100%+	TSMC production pace, DC deployment decisions

11.8　Regional Market Structure Forecast

North America (~40% of global consumption by 2030E): AI hyperscale core; Meta/Google/Microsoft/Amazon/OpenAI.

Asia-Pacific (~35%): China domestic AI buildout accelerating; Southeast Asia (Indonesia, Vietnam, Thailand) digital infrastructure buildout.

Europe (~12%–15%): AWS/Microsoft/Google European DC expansions; NIS2 compliance manageable.

Middle East/Africa/LatAm (~10%): High growth; primarily telecom-side (FTTH/5G) demand.

11.9　Speed-Tier Quantitative Forecast

400G: ~50–60 million units/year in 2024; ~30–40 million units/year by 2030 (replacement + mid-low applications); ASP declining to <USD 50.

800G: ~7.5 million units (2024) → >20 million (2025) → peak ~30–40 million/year (2026–2027) → gradual decline as 1.6T substitutes.

1.6T: ~3–5 million units (2025) → >10 million (2026) → ~30–50 million (2027–2028); potentially highest-volume speed tier by 2029–2030 if CPO slower than expected.

3.2T/CPO: Starting from 2027; <several million units 2030; high ASP (USD 500–2,000/port) amplifies revenue contribution.

11.10　China Policy Support for Five-Year Development

"15th Five-Year Plan" (2026–2030): 1) Compute network construction acceleration (national compute target >>300 EFLOPS by 2030); 2) Optical chip "innovation challenge" national projects (EML, InP, high-speed DSP)—estimated >CNY 10 billion state investment 2026–2030; 3) STAR Market support for specialty optical chip companies; 4) 5G-A and 6G pre-research driving new telecom-side demand; 5) "Digital Silk Road" submarine cable investments.

11.11　Is There a Ceiling on "Chinafication" of the Global Optical Transceiver Market?

Pure market logic would drive Chinese share toward ~100% (solar panel analogy). Reality: 1) Technology moats (coherent modules, high-end InP/EML/DSP) constrain shares in premium segments; 2) Customer supply-chain diversification pressure; 3) Geopolitical intervention tail risk. Report baseline: Chinese vendor market share ceiling ~60%–65% by 2030—above current 50% but well below a solar-panel-style monopoly.

11.12　Next Ten Years: 2026–2036 Outlook

2010–2026: From a follower in low-speed access modules to global leader with 7-of-10 top-ten share. 2026–2036: The core narrative shifts from "how to manufacture" to "how to innovate"—from producing the world's most pluggable transceivers to leading CPO and photonic integration standard-setting; from purchasing foreign EML and DSP to becoming a leading global optical chip supplier. The core challenge: optical communication's future high ground (photonic integration, CPO, advanced-node DSP) intersects deeply with semiconductor and advanced manufacturing domains where China still has gaps (EDA, advanced process nodes, lithography). Solving those gaps is the central challenge for China's entire technology manufacturing ecosystem in the next decade.

Chapter 12　Conclusions

12.1　Core Findings Summary

Finding 1: AI compute is the historic tailwind of this transceiver generation. The 2024 global transceiver market (~USD 17–18 billion, +52%) reflects the largest single technology infrastructure investment wave in history. Unlike prior 3G/4G/5G cycle-driven builds, this demand is structural, continuous, and exponential—from the insatiable compute requirements of large model training and inference. AI capex growth of 20%–35% per year through 2027 provides the most certain transceiver demand window in industry history.

Finding 2: China's global dominance in transceiver module manufacturing is now established. Seven of the top ten global optical transceiver suppliers are Chinese; CR3 combined global share exceeds 50%. This is "global leadership"—not merely "in the top three." Innolight's 2024 revenue of CNY 23.862 billion (+122.64%) and net profit of CNY 5.171 billion (+137.93%) are the most direct financial expression of this structural reality. The formation of this position took a decade of precision manufacturing accumulation, scale economics, technology iteration velocity, and deep customer binding—not easily dismantled in the short term.

Finding 3: Upstream optical chips are the next competitive battlefield. The China module manufacturers' competitiveness is validated. The next five years belong to YJSemi (688498) and Changelight (688048)—EML/CW breakthrough narratives will be the core story of China's optical communications industry upgrade. The InP substrate domestic rate below 15% and high-end 1.6T EML domestic rate below 20% represent both risk (supply chain dependency) and opportunity (large domestic substitution market). Once national-domestic EML scales from 100G to 200G breakthrough, China's transceiver supply-chain self-sufficiency will undergo a qualitative transformation.

Finding 4: CPO is a distant concern, not an immediate threat—but leaders must begin preparation. CPO penetration remains below 5% of new DC optical interconnects in 2026; the pluggable window extends at minimum to 2027–2028. Chinese leaders' CPO technical reserves (Eoptolink's TFLN platform, Innolight's stated CPO R&D, ZTE Micro and Huawei photonics programs) provide foundation for competing in the 2028+ CPO era. Companies that simultaneously maximize pluggable shipments through 2027 and complete CPO technical preparation will be best positioned in the post-CPO market restructuring.

Finding 5: This is one of China's most certain strategic advantages in the AI era. Unlike AI chips (sanctioned/technology-lagging) and semiconductor equipment (import-heavy), optical transceivers are a genuine Chinese advantage node in the global AI infrastructure supply chain—one where China is the displacer, not the displaced, and where restricting China's supply would severely damage US tech companies' competitive position. In a geopolitical environment where finding "China as clear advantage party in a massive-demand intersection" is rare, optical transceivers are one of the few credible answers.

12.2　Risk Caveats

This report's conclusions are based on the baseline scenario. Significant deviations will result from: AI capex pace disappointing (largest tail risk); US export controls extending to optical transceiver finished goods or upstream DSP/EML chips (low probability, extreme impact); CPO commercializing faster than expected (long-term structural risk, requires ongoing monitoring); internal Chinese transceiver price competition exceeding expectations (cyclical risk; heads benefit).

12.3　Factory Data Platform Application Note

For verifying the breadth and depth of China's optical communications supply chain, the platform at www.tianxiagongchang.com, as a B2B factory database covering 4.8 million active factories, provides comprehensive factory identification across optical transceiver manufacturers, optical component makers, optical chip assembly/test firms, and fiber cable manufacturers. Unlike other business-lookup platforms, its value lies in precisely distinguishing "verified active factories" from "trading companies/shells"—a capability of irreplaceable significance for supply-chain tracing, capacity verification, and competitive landscape assessment in optical communications—an industry where genuine manufacturing creates genuine value. Procurement teams and researchers in optical transceiver, optical component, and optical chip domains can use the platform's factory search to directly locate verified manufacturing entities in Wuhan Optics Valley, Suzhou, and Chengdu clusters, enabling precision supply-chain management.

12.4　Research Summary Table

Dimension	Core Data and Finding
Global market	2024: ~USD 17–18 bn (+52%); 2030E: ~USD 40–50 bn (CAGR 15%–22%)
China position	7 of top 10; CR3 global share >50%
Innolight	2024 revenue CNY 23.862 bn (+122.64%); global #1 (~22% share)
Eoptolink	2024 revenue CNY 8.647 bn (+179.15%); net profit +312.26%; fastest growth
800G status	2024 ~7.5M units shipped (+6× YoY); 2025E >20M units
1.6T progress	Mass-production ramp; China H1 2025 ~45% global share; 2026 demand >10M units
Optical chip gaps	InP substrate domestic rate <15%; high-end EML <20%; DSP nearly 100% foreign
Core risks	AI capex pace (largest); supply-chain tail risk; CPO long-term variable
5-year CAGR	Global market 15%–22%; Chinese vendors 20%–28%
Industrial clusters	Wuhan Optics Valley (core) + Suzhou (high-speed modules) + Chengdu (emerging pole)
Technology direction	800G→1.6T→3.2T/CPO; SiPh/TFLN penetration rising
Strategic positioning	Most certain Chinese advantage segment in AI compute downstream chain; optical chip domestic substitution is the next decade's theme

Data in this report is current as of H1 2025; listed company financial data is based on publicly disclosed 2024 annual reports. Forecasts carry uncertainty and are for research reference only; they do not constitute investment advice.

12.5　Optical Communications and China's Technology Competitiveness: A Broader Narrative

Positioning optical transceivers in the broader Chinese technology competitive landscape reveals strategic significance beyond industry alone.

Over the past decade, China's technology narrative has focused on shortfalls—no advanced chip manufacturing, equipment sanctioned, EDA tools foreign-dominated, high-end materials imported. These shortfalls are real. Optical transceivers are a rare "Chinese long-board" story: here, China is the leader, not the follower; the one causing dependency, not the one experiencing it.

This "leader" strategic value matters in geopolitical competition just as much as the shortfalls: the cost of constraining China in optical communications is extreme, because global AI infrastructure's main participants (Meta, Google, Microsoft, Amazon) are deeply dependent on Chinese transceiver products. Even if the US government intended to restrict supplies, it would face the dilemma of directly harming its own tech companies' competitiveness.

This is why the optical transceiver industry represents not merely a commercial track success, but a microcosm of China's "concentrate advantages, achieve irreplaceable global positions in key tracks" AI-era technology strategy.

12.6　Framework for Investors and Researchers

Short-term tracking (quarterly): AI hyperscaler capex announcements (quarterly earnings); major transceiver companies' quarterly results (revenue growth, gross margins, overseas revenue share); 800G/1.6T transceiver quarterly shipments (LightCounting reports, earnings calls).

Medium-term indicators (annual): 1.6T production ramp and Chinese vendor market share evolution; CPO commercialization real progress (TSMC production scale, Broadcom/NVIDIA announcements); 100G EML domestic substitution progress (YJSemi, Changelight annual report product shipments).

Long-term structural variables (3–5 years): DSP chip domestic substitution path (Marvell/Broadcom China policy, LPO commercial scale); InP substrate domestic breakthrough (domestic InP companies achieving 3–4-inch production); CPO disruption timeline (whether 2028 pre-date hyperscale acceleration occurs); next-generation Chinese AI large-model training cluster optical interconnect procurement patterns.

12.7　Global Optical Communications Research Resource Index

Market research: LightCounting (monthly transceiver shipment tracking), Dell'Oro Group, Ovum/Omdia, IDC.

Conferences: OFC (March, US—global premier), CIOE (September, Shenzhen—China largest), ECOC (September, Europe—more research-oriented).

Industry media: Lightwave Online, Telecom Ramblings, EETimes Optical; Chinese: OFweek Optical (fiber.ofweek.com), C114 Communications (c114.com.cn), Xunshi Optical (iccsz.com).

Investor relations: Innolight (ir.innolight.com), Eoptolink (300502.sz), HGtech Photonics (002281.sz) on A-share disclosure platform; Coherent Corp (ir.coherent.com), Lumentum (ir.lumentum.com) English IR sites.

12.8　Research Limitations and Future Research Directions

Limitations: Financial data based on public annual reports (through 2024); market size estimates from multiple agencies with differing definitions; detailed product-level revenue breakdowns for YJSemi and Changelight not publicly disclosed; hyperscale customer-specific transceiver purchase volumes are commercial secrets; 1.6T LPO deployment penetration specifics not publicly disclosed by vendors or customers.

Future research directions: CPO TCO economic modeling (detailed total-cost-of-ownership comparison); China optical chip catch-up path quantitative research (patent analysis, product roadmaps, capital investment → domestic substitution timeline); geopolitical scenario analysis (mild/moderate/extreme export control impact on Chinese transceiver industry output, exports, employment).

This report will be updated in 2027 to reflect 1.6T mass production, CPO commercialization, and optical chip domestic breakthrough latest developments.

Data Sources

The factory entity identification and in-production verification in this report is based on the factory database at www.tianxiagongchang.com; industry data is synthesized from the following public sources and cross-verified:

• Market research institutions and associations: LightCounting, Ovum/Omdia, Dell'Oro Group, IDC, MarketsandMarkets, Zhiyan Consulting, Huajing Industry Research, Qianzhan Industry Research, CAICT (China Academy of Information and Communications Technology)—Information Photonics Technology Development and Application Research Report (December 2024), Zhongzheng Pengyuan Credit Research "Optical Transceivers: Welcoming 1.6T Era Under AI Compute Drive" (September 2024)
• Listed company annual reports and announcements: Innolight (300308) 2024 Annual Report; Eoptolink (300502) 2024 Annual Report; HGtech Photonics (002281) 2024 Annual Report; HGtech Science & Technology (000988) Q1–Q3 2024 Report; Accelink Technology (300394) 2024 Annual Report; Cambridge Industries Group (603083) 2024 Annual Report; YJSemi (688498) 2024 Annual Report; Changelight (688048) 2024 related announcements
• Overseas public company financials: Coherent Corp (COHR) FY2024 Annual Report; Lumentum Holdings (LITE) FY2024 Annual Report; Marvell Technology (MRVL) FY2025/Q4 Earnings; Broadcom Inc. (AVGO) FY2024 Annual Report
• Policy documents: "East-Data-West-Computing" Engineering (NDRC/MII, 2021–2025); "14th Five-Year" Information and Communications Development Plan; "15th Five-Year" Planning Guidance (November 2025, compute network/digital economy sections); Hubei Province Optoelectronic Information Industry Trillion-RMB Cluster Development Plan (2024); CAICT Computing Infrastructure Development Index (2024)
• Industry media and research reports: C114 Communications (c114.com.cn), Xunshi Optical (iccsz.com), OFweek Optical (fiber.ofweek.com), Securities Market Weekly, Sina Finance, Jiemian News, Economic Observer, Securities Times; multiple broker industry deep-dives (Dongwu Securities, Huatai Securities, CITIC Securities, Ping An Securities, Guotai Junan); 36Kr "Decoding the Origin Story of Optical Transceiver 'Three Musketeers'" (2024); IDC CPO Technology Report CHC52899425 (July 2025)

Note: Different institutions' methodologies create definitional divergences for the same metric (e.g., global 2024 transceiver market USD 17–18 billion = comprehensive datacom/telecom/access scope; USD 11.9 billion = optical transceiver-only scope; China InP substrate domestic rate <15% = full 2-inch+ specification scope). This report annotates major divergences or presents range estimates. Future-oriented figures carry uncertainty; they are for research reference only and do not constitute investment advice.