China’s Two-Decade Optical-Communications Drive Captures a Critical Chokepoint in Global AI Data Centers

Two decades of state procurement built a mass-production base for optical communications
China commands 56% of global outsourced optical-module manufacturing capacity through cost and delivery advantages
AI data-center expansion amplifies China’s grip over the supply chain

Optical cables laid across China over more than two decades have returned as strategic assets in the AI era. Chinese companies, having secured economies of scale and cost competitiveness through massive procurement by state-owned telecommunications carriers, are expected to account for 56% of global outsourced optical-module manufacturing capacity this year. As optical links connecting tens of thousands of graphics processing units (GPUs) determine the throughput and uptime of AI data centers, China’s dominance over the supply chain for optical modules and cables is strengthening further.

56% of Global Output, China’s Dominance

According to the Hong Kong-based South China Morning Post (SCMP) on August 12, China’s optical-communications industry is rapidly expanding its influence as competition over AI data-center performance shifts from computing capacity to data-transfer speeds. Market research firm TrendForce estimates that Chinese optical-module suppliers will account for 56% of global outsourced manufacturing capacity this year. Chinese suppliers serving the North American market alone, including Zhongji Innolight and Eoptolink, represent 46% of that capacity. China’s command of optical-module assembly, packaging, and testing has positioned the country as a pivotal pillar of the global AI-infrastructure supply chain.

Optical communications use light waves as carrier signals and optical fiber or free space as the transmission medium for information. The technology offers ultra-high bandwidth, extremely low transmission loss, strong resistance to electromagnetic interference, and robust security. As digitalization and intelligent systems advance, optical communications have become foundational infrastructure for the information society, with applications spanning a broad and continuously expanding range of fields.

Technological advancement is coinciding with rapid earnings growth. Innolight has already begun mass production and expanded shipments of next-generation 1.6-terabit optical modules, while developing 3.2-terabit products and pursuing customer qualification for near-packaged optics (NPO). Its first-quarter revenue rose 192.1% from a year earlier, while net profit attributable to shareholders surged 262.3%. The company also raised HK$53.41 billion, or approximately $6.8 billion, through its Hong Kong listing last month, accelerating investment in research and development (R&D) and overseas capacity expansion.

Cost Competitiveness Built by State Procurement

China’s investment in optical communications began with efforts to expand telecommunications access in the 1990s. Its vast territory, low fixed-line penetration, and the divide between coastal and inland regions required enormous spending to build long-haul backbone networks and local access systems simultaneously. The Chinese government expanded nationwide networks by combining investment in railways, broadcasting networks, and state-owned telecommunications carriers, while demand for optical fiber shifted to large-scale procurement in the early 2000s. Long-term orders increased manufacturers’ plant utilization rates and generated learning effects that improved production yields and cost competitiveness.

A symbolic project was CNCnet, the high-speed optical backbone launched by China Netcom in 1999. China Netcom, jointly established by the Chinese Academy of Sciences, the State Administration of Radio, Film and Television, the Ministry of Railways, and the Shanghai municipal government, built a network spanning 22,700 miles and connecting 126 cities. The network was designed with transmission capacity of 40 gigabits per second (Gbps), with construction scheduled over 30 months. This period also established the template for China’s industrial policy, which bundled nationwide network construction, localization of telecommunications equipment, and carrier competition into a unified initiative.

The “Broadband China” strategy introduced by the State Council in 2013 elevated optical-communications investment into a nationwide industrial policy. Fiscal resources flowed into urban fiber-access-network expansion and deployment across central, western, and rural regions, sustaining orders from China’s three major telecommunications carriers. According to the Ministry of Industry and Information Technology, China’s total optical-cable length reached 74.99 million kilometers at the end of last year. Fixed broadband ports totaled 1.251 billion, while fiber-to-the-home/office (FTTH/O) ports reached 1.21 billion. FTTH/O accounted for 96.8% of the total. Optical cable additions in 2025 alone reached 2.113 million kilometers.

Overseas Production Accelerates After Domestic Dominance

Concentrated procurement by state-owned telecommunications carriers gave manufacturers a foundation of long-term orders and mass production. As China Mobile, China Telecom, and China Unicom set nationwide network specifications and moved to bulk purchasing, supply chains became increasingly dense across optical-fiber preforms, fiber drawing, cable assembly, and connectors. According to Yangtze Optical Fibre and Cable (YOFC), China Mobile ordered 3.14 million fiber-kilometers (fkm) of ultra-low-loss G.654.E optical fiber last year, up 156% from the previous year. China Telecom procured 2.24 million fkm during the same year.

As economies of scale accumulated, overseas localization of manufacturing hubs also accelerated. YOFC operates eight production facilities across six countries: Indonesia, South Africa, Brazil, Poland, Germany, and Mexico. Overseas revenue rose 47.8% year on year to RMB6.092 billion, or approximately $850 million, last year, lifting its share of total revenue to 42.7%. Global optical-cable demand in the same year reached 549 million kilometers in fiber-length terms, according to commodities and industrial-market research firm CRU, representing growth of 3.9%. In the patch-cable market, where product specifications are highly standardized and procurement is sensitive to price and delivery lead times, mass-production capacity and dense distribution networks determine purchasing competitiveness. These advantages in manufacturing and distribution have expanded the reach of Chinese cables into data centers and enterprise networks worldwide.

Table 1. Failure Factors and Availability Risks in AI Data-Center Optical Links

Optical Connectivity Determines AI Computing Efficiency

The global manufacturing and distribution networks built by Chinese companies have entered a new growth phase alongside expanding AI data-center investment. During large-scale training, thousands to tens of thousands of GPUs exchange parameters and intermediate computational values at every step. Link latency or packet loss leaves expensive accelerators idle and reduces the efficiency of additional computing equipment. Copper connections, long used to link servers and switches, suffer greater signal loss and heat generation as transmission speeds rise, while their viable reach shortens sharply. Intel has indicated that high-speed electrical input/output has a practical transmission distance of less than one meter. Its optical input/output chiplet prototype, unveiled in 2024, transmitted 4 terabits per second (Tbps) bidirectionally over distances of up to 100 meters, consuming 5 picojoules (pJ) per bit—one-third the energy used by conventional pluggable optical modules.

Copper’s transmission-distance constraints have moved the application of optical connectivity from inter-data-center and inter-rack communications into rack interiors and areas surrounding circuit boards. AMD, Broadcom, Meta, Microsoft (MS), Nvidia, and OpenAI launched the Optical Compute Interconnect Multi-Source Agreement (OCI MSA) in March this year. The OCI MSA specifies 200Gbps per direction in its first-generation standard and bidirectional transmission of up to 800Gbps per fiber in the second generation. The target for follow-on standards is at least 3.2Tbps per fiber. It supports pluggable optical modules, board-mounted optics, and co-packaged optics (CPO), placing the shift from electrical wiring to optical links inside GPU racks within the scope of standardization.

As optical connectivity moves into rack interiors, reliability has emerged as a variable governing GPU utilization rates. Optical links form an integrated transmission system in which cables, connectors, transceiver modules, lasers, and optoelectronic conversion components operate sequentially. Cable cuts, connector contamination, deterioration of modules and lasers, and overheating of conversion components have long been identified as primary causes of signal degradation and link failures. Microsoft’s tracking of thousands of optical channels in a large backbone network for more than a year in 2016 also found significant variation in link-level availability, while signal-quality deterioration was confirmed as a leading indicator of impending failures. A failure in a single optical link can delay synchronization among GPUs and accumulate waiting time, reducing throughput across the entire cluster. Transmission capacity, failure frequency, and recovery time therefore jointly determine network availability at AI data centers.

The Optical-Communications Supply Chain, the Next Battleground After GPUs

As transmission bottlenecks and failure risks rise, global technology companies are accelerating optical-equipment upgrades and mergers and acquisitions (M&A). Nvidia unveiled Spectrum-X and Quantum-X Photonics switches last year, each capable of handling 1.6Tbps per port. The company said the systems deliver 3.5 times greater energy efficiency and 10 times stronger network resilience than conventional pluggable optical modules. Nokia completed its acquisition of Infinera in February of the same year, securing a North American hyperscale customer base and optical-transmission technology. The acquisition also included Infinera’s in-house indium phosphide (InP) photonic integrated circuits and optical-transmission equipment business.

Competition to secure supply chains has also spread to long-term purchase agreements. Meta signed a supply agreement with Corning in January this year for optical fiber, cables, and connectivity equipment worth up to $6 billion. Corning will expand its optical-cable plant in Hickory, North Carolina, and designate Meta as a key customer of the new facility. Nvidia also invested $2 billion each in Lumentum and Coherent in March, for a combined $4 billion. Its agreements with the two companies include multibillion-dollar long-term purchase commitments and rights to secure production volumes of advanced lasers and optical-networking products.