The latest report from JPMorgan has ignited the market— the CPO (Co-Packaged Optics) market is expected to exceed $1 billion by 2028 and reach $5 billion by 2030! NVIDIA’s GB200 copper cable solution is being implemented, TSMC has made breakthroughs in optical communication technology, and global AI server shipments surged by 35% in Q2… The infrastructure for computing power is deeply spreading from chips to transmission layer hardware.
Three Core Tracks Driven by Technology
1. CPO (Co-Packaged Optics)
Essentially, it integrates chips and optical modules into a single unit (reducing power consumption by 40% and increasing speed by 50%). The density of computing power in AI clusters is surging, while traditional pluggable optical modules are approaching physical limits.
2. High-Speed Copper Cable Connections
Essentially, it is the “golden artery” for short-distance data transmission in data centers (costing only 1/3 of optical cables).
Catalytic event: NVIDIA’s GB200 adopts a copper cable solution, increasing usage per cabinet by 4 times.
3. High-End PCBs
Essentially, they are the “nervous system” of computing devices (carrying interconnections between chips/components).
Catalyst: The number of layers in server PCBs has increased from 12 to over 20, doubling the value per unit.
Overview of 10 Core Companies
Dissecting the Core Logic of the Industry
1. Accelerated Technology Iteration
CPO penetration rate: 3% in 2025 → 45% in 2030 (LightCounting data)
Copper cable reuse: The amount of copper cable used in NVIDIA’s GB200 per cabinet reaches 2000 meters (a 400% increase compared to H100).
2. Domestic Substitution in Deep Waters
High-end PCB materials: Japan’s Ajinomoto ABF film has a market share of 99% → Huazheng/Shengyi are accelerating breakthroughs.
224G connectors: Dominated by overseas giants → Dingtong Technology has been certified by Huawei.
3. Increased Cost Sensitivity
The copper cable solution reduces wiring costs per cabinet by 62%.
CPO reduces the power consumption cost of optical modules by 50%, becoming a necessity for large-scale clusters.
In Conclusion:
As the competition for computing power shifts from “chip battles” to “transmission battles,” the cables, circuit boards, and optical devices hidden in cabinets are becoming the “new ammunition” in the AI arms race. However, the speed of technological iteration may be faster than expected— the true moat for enterprises lies in their ability to keep pace with NVIDIA’s next-generation architecture transitions.
(Note: This article objectively analyzes industry trends, and the information about the companies involved is sourced from public data. There is uncertainty in technological development, and the market carries risks; decisions should be made independently.)