News
Optical modules serve as the core carriers for interconnecting data center network devices, such as switches, routers, and servers. Their importance is reflected in three key aspects:
As the computing power of CPUs and GPUs grows exponentially, network interconnection must keep pace; otherwise, it will become a bottleneck for the entire system. The evolution of optical module rates directly determines the smoothness of the "arteries" within data centers.
Large-scale data centers house tens of thousands of optical modules, and their power consumption and physical size directly affect the data center’s operational costs (OPEX) and rack space utilization. Optical modules with higher rates, lower power consumption, and higher density are the core to reducing costs and increasing efficiency.
The technical selection of optical modules determines the flexibility and reliability of the network architecture. The trade-off between the plug-and-play design of pluggable modules and the emerging Co-Packaged Optics (CPO) technology is a strategic decision that data center operators must face.
Current Status: Large-scale deployment has been achieved across the industry, serving as the backbone of the spine-leaf architecture in hyperscale data centers. Meanwhile, it is being popularized in small and medium-sized enterprise data centers and edge computing nodes. Adopting the QSFP-DD form factor, it boasts a mature supply chain and significant cost-performance advantages.
Implementation Method: Mainly based on 4X100G wavelengths or 8x50G fiber channels, the technical solution is stable and reliable, compatible with most existing network equipment.
Current Status: 2025 marks the first year of large-scale popularization of 800G optical modules. No longer limited to leading AI training clusters, they have been widely applied in scenarios such as backbone networks of medium and large cloud service providers, High-Performance Computing (HPC) centers, and AI inference nodes, becoming the core choice to meet the high-density interconnection needs of GPU clusters.
Implementation Method: The mainstream adopts 8x100G channels, and the QSFP-DD form factor continues to dominate the market. Some manufacturers have launched more compact packaging solutions, further optimizing compatibility with 400G infrastructure and lowering the upgrade threshold.
Outlook: It entered the initial commercial stage in 2025, with leading cloud service providers and supercomputing centers starting small-batch pilot deployments, and large-scale volume is expected in 2026. It retains the pluggable module form, with OSFP-XD/OSFP-HS form factors becoming the mainstream choice.
Challenges: Core challenges focus on cost control and power consumption optimization. Through the large-scale application of silicon photonics technology and advanced DSP chips, the goal of "doubling the rate without increasing power consumption" is gradually achieved.
Outlook: In 2025, many manufacturers completed the development and performance testing of 3.2T optical module prototypes and began joint verification with leading data center customers, with initial commercialization expected in 2027. 3.2T will become the technical limit of pluggable optical modules, posing extreme challenges to SerDes rate, chip power consumption, and heat dissipation.
Form Factor: The traditional pluggable architecture is approaching its performance limit at the 3.2T rate. Co-Packaged Optics (CPO) technology achieved key technological breakthroughs in 2025, becoming the core candidate solution for 3.2T and higher rates.
Rate improvement is not a simple numbers game; it is backed by the integrated innovation of a series of cutting-edge technologies:
PAM4 (Four-Level Pulse Amplitude Modulation) has become the standard technology for 800G and higher rates. It was further optimized in 2025, with continuous improvements in anti-interference capability and transmission efficiency, laying the foundation for the commercialization of 1.6T.
In 2025, DSP chip performance achieved a leap-forward improvement, supporting more complex signal compensation algorithms while reducing power consumption by more than 20%, becoming the core support for the commercialization of 1.6T optical modules.
InP: It remains the core solution for high-performance optical modules, maintaining performance advantages at the 1.6T rate. Costs are gradually decreasing through large-scale production.
SiPh: It entered the large-scale application stage in 2025, with the shipment proportion of 800G silicon photonics optical modules exceeding 30%. With its advantages of high integration and low cost, it has become the key technical path for cost reduction and efficiency improvement of 1.6T optical modules.
In 2025, the SerDes channel rate of switch ASIC chips fully entered the 224G era, and some manufacturers have launched 320G SerDes prototypes, providing underlying support for 1.6T (8x200G) and 3.2T (16x200G) optical modules.
