Core Insights - CPO has emerged as a highly discussed technology in AI data centers, positioned as a potential solution to bandwidth, latency, and power consumption challenges faced by AI workloads [2][4] - Despite impressive demonstrations at industry events, actual deployment of CPO in most data centers remains limited, with a cautious attitude prevalent among data center managers [2][4] - The transition to CPO is not just a technical upgrade but involves significant changes in user psychology, risk tolerance, and the relationship between infrastructure buyers and suppliers [2][4] Group 1: CPO's Relevance and Advantages - CPO is not a new technology; its concept dates back years, but its relevance has surged with the rise of AI as a core workload in data centers [4] - CPO is seen as a solution to specific problems in AI data centers, particularly as the industry moves towards higher speeds (112G and 224G SerDes) where electrical interconnects become inefficient [4] - The advantages of CPO include energy efficiency improvements of 20%-40% in interconnect power consumption at 800G to 1.6T rates, potentially saving 5-15 watts per 800G port and reducing overall switch power consumption by 200-500 watts [5] Group 2: User Perspectives and Adoption Challenges - There is a divide in attitudes towards CPO among data center operators, with some fascinated by its potential while others are skeptical about operational risks [8] - Many traditional data center managers have limited knowledge of CPO, and even established suppliers of pluggable optical modules are often unaware of its implications [8] - CPO is viewed as a critical technology for ultra-large-scale enterprises, which are already experimenting with it to manage power budgets and enhance performance [9] Group 3: Transition Technologies and Market Dynamics - Users are likely to adopt transitional solutions like NPO and LPO before fully committing to CPO, as these provide some benefits without the risks associated with full co-packaging [12][13] - The rise of external laser sources (ELSFP) may reintroduce some level of modularity to CPO systems, addressing maintenance concerns while also presenting new risks [14] - The influence of key suppliers like Broadcom and NVIDIA is significant in shaping user perceptions and driving CPO adoption through credible engineering support [15] Group 4: Future Deployment Phases - The deployment of CPO is expected to follow a three-phase approach: skepticism to acceptance (2026-2028), acceptance to reliance (2029-2032), and reliance to optimization (2032-2035) [23] - The success of CPO will depend on overcoming challenges related to laser technology, packaging yield, and thermal design [23] - The evolution of AI workloads and the ongoing development of copper cabling will also impact the adoption rate of CPO technology [24]

Group 1 - Ren Zhengfei stated that China has opportunities in mid-to-low-end chips, with numerous domestic chip companies making efforts [1] - The transition to large model era is driving the demand for enhanced interconnectivity between computing cards, leading to an upward trend in domestic computing power construction [1] - Dell'Oro Group predicts that by 2025, the adoption rate of 800G switch ports is expected to exceed 400G data center switch ports, accounting for over 25% of data center switch ports [1] Group 2 - The global sales volume of CPO switches is projected to grow from 50,000 ports in 2023 to 4.5 million ports by 2027 [2] - OCS all-optical network switches are expected to address bandwidth latency and dynamic topology issues in large cluster training across cabinets and regions [2] Group 3 - Leading companies in the switch sector include Ruijie Networks and Fenghuo Communication [3] - Key players in the copper cable market are Huafeng Technology and Wolong Nuclear Materials [3] - Notable companies in the optical module sector include Zhongji Xuchuang, Xinyi Sheng, and Tianfu Communication [3]