In 2026, the computing power sector is entirely focused on GPUs. 100,000-GPU clusters are being built one after another, large model training is in full swing, and chip supply schedules have become the most closely watched topic in the industry. However, an increasing number of industry players have noticed that as computing infrastructure construction enters the critical phase of networking and delivery, the underlying Optical Transport Network (OTN) connecting all computing nodes has already come under pressure. From extended lead times for indium phosphide (InP) substrates to accelerated production scheduling for high-speed optical modules, and tightening turnkey delivery schedules for high-end OTN equipment, upstream tensions are cascading down the industry chain. Notably, the signals emerging across the industrial chain are mixed. Recently, Dongshan Precision stated in investor communications that as capacity continues to expand, optical module prices will exhibit a controllable and reasonable downward trend in the future. Zhongji InnoLight also publicly stated that 1.6T product prices will remain at a reasonable level based on market supply and demand, without drastic fluctuations. This indicates that the optical communication industry is gradually transitioning from rapid expansion to a more rational rebalancing of supply and demand. Amid this dynamic adjustment, the delivery and turnkey assurance capabilities of OTN equipment deserve more attention than short-term price fluctuations.
01. In 2026, Why Is OTN Suddenly in the Spotlight?
Rewinding two or three years, few would place OTN at the center of AI industry discussions. Back then, the industry focused on 5G bearer networks and metropolitan area network (MAN) upgrades for operators, with OTN rarely capturing the attention of capital markets. The shift originates first at the national strategic level. In 2026, the Outline of the 15th Five-Year Plan explicitly proposed advancing the "East Data, West Computing" project to build a multi-tiered computing infrastructure system and a national integrated computing network. The central government introduced the "Six Networks" construction initiative, integrating the computing network into the national strategic infrastructure lineup for the first time, alongside water and new power grids. According to estimates by relevant institutions, computing network investment in 2026 will reach the 4 trillion RMB level. If national strategy dictates the direction, the rapid evolution of the AI industry has triggered an early explosion of this demand. By March 2026, the national intelligent computing capacity reached 1.882 million PFLOPS (FP16), 2.5 times the figure from the same period last year. A total of 42 10,000-GPU intelligent computing clusters and over 70 major computing corridors have been established, with interconnection latency between the eight major hubs compressed to under 20 milliseconds. In July 2026, China's first 100,000-GPU super-intelligent converged computing resource pool was officially commissioned in Zhengzhou. As computing infrastructure construction enters an intensive deployment phase, an increasing number of large model training tasks require collaboration across multiple intelligent computing centers, while inference workloads demand flexible scheduling across different regions. The AI industry must build a network capable of connecting national computing resources while also addressing convenient computing access for end users. This is the underlying rationale behind the Ministry of Industry and Information Technology's (MIIT) "Millisecond Computing" metropolitan initiative: interconnection latency between computing centers must be under 1 millisecond, and latency for key locations to access computing resources must also be below 1 millisecond. As network requirements evolve from mere connectivity to millisecond-level coordination, the network itself has transitioned from a communication infrastructure to a core component of computing infrastructure. Against this backdrop, OTN has assumed the critical responsibility of network support in the AI era.
02. In the AI Era, What Is Redefined Is Not OTN, But the Role of the Network
Over the past two decades, communication networks have focused on connecting people. In the AI era, the network faces the challenge of connecting computing power. Large model training is essentially ultra-large-scale parallel computing, requiring tens of thousands of GPUs to participate simultaneously, with gradient parameter synchronization completed within extremely short intervals after each computation round. To use a vivid analogy: AI training is not a 100-meter sprint but a multi-person three-legged race. The final speed is determined not by the fastest runner, but by the slowest. For a 100,000-GPU cluster, even microsecond-level network jitter can cause the capacity utilization rate of the entire cluster to drop by several or even ten percentage points. OTN has returned to the center stage in this transformation due to its irreplaceable technical characteristics. Unlike the statistical multiplexing mechanism of traditional IP networks, OTN provides physical-layer hard pipes with dedicated bandwidth, deterministic latency, and zero packet loss transmission. Under the strict constraints of millisecond computing metrics, OTN is virtually the only technical option capable of achieving millisecond-level interconnection between computing centers. Even more noteworthy is the comprehensive upgrade of the AI-OTN architecture. Ao Li, Chief Engineer of the China Academy of Information and Communications Technology (CAICT), explained during MWC Shanghai 2026 that the AI-OTN architecture delivers deterministic experiences for rapidly evolving AI applications through the deep integration of "computing" and "networking." Next-generation OTN equipment is deeply embedded with computing-aware capabilities, enabling real-time identification of differentiated service requirements for training, inference and storage synchronization traffic. Through coordinated computing-network scheduling, it dynamically allocates optimal optical path resources for each service type, upgrading OTN from a passive optical transmission pipe to an active computing transmission brain. The outcomes of this technological leap have been validated at the market level. Between May and June 2026, China Mobile launched its centralized procurement of CPE OTN equipment for 2026–2027. China Telecom conducted a centralized procurement for 100G (200G) DWDM equipment expansion covering 29 provinces, supporting OTN 2.0 flexible hard-pipe technology. Meanwhile, Dell'Oro Group predicts that the global optical transport equipment market will grow by 16% in 2026, exceeding $18 billion in scale. The strategic value of OTN lies in its role as the tangible physical foundation of the computing network, the core enabler for operators transitioning to computing-as-a-service, and an inseparable component of computing security and autonomous control. Through its three-fold evolution—AI-driven architecture, ultra-broadband ports, and