Sinolink Securities: Large-Scale AI Clusters Accelerate CPO Adoption

Surging AI Computing Demand Presents Historic Opportunity for CPO Technology

As demand for large model training and inference experiences explosive growth, AI infrastructure faces unprecedented challenges. Sinolink Securities recently published a research report indicating that large-scale AI clusters are driving the accelerated deployment of CPO (Co-Packaged Optics) technology, with core companies across the supply chain poised to seize major development opportunities.

The report emphasizes that AI development is driving the continuous escalation of computing power and network demands along two dimensions. On one hand, training cluster scales continue to expand, with the number of GPUs required for a single training task evolving from several thousand to tens of thousands or more. On the other hand, as large model applications become more widespread on the inference side, token consumption continues to grow, making computing demand equally significant.

CPO Technology: Solving Three Major Pain Points of Large-Scale Clusters

In traditional data center architectures, optical modules exist in a "pluggable" form. As data transmission rates advance from 400G to 800G and even 1.6T, bottlenecks in power consumption, density, and signal integrity with conventional approaches have become increasingly apparent.

Sinolink Securities believes CPO technology can effectively address three core pain points facing large-scale AI clusters:

First, reducing power consumption. CPO packages the optical engine directly with the switch chip, dramatically shortening electrical signal transmission distances and significantly lowering energy consumption. In ultra-large-scale clusters with tens of thousands of GPUs, power savings are critically important — affecting not only operating costs but also data center cooling and power supply capabilities.

Second, increasing interconnect density. As AI cluster scales expand, switches need to provide more ports and higher bandwidth. CPO solutions integrate optical components inside the package, freeing up substantial panel space and enabling a single switch to support higher interconnect density, thereby meeting the urgent demand for high-bandwidth interconnection in large-scale clusters.

Third, improving high-speed transmission stability. At 800G and above, electrical signal transmission losses on PCBs increase sharply, posing severe challenges to signal integrity. CPO moves the electro-optical conversion point forward into the chip package, greatly shortening the high-speed electrical signal transmission path and fundamentally improving signal quality and transmission stability.

Supply and Demand Convergence Signals Imminent CPO Acceleration

Sinolink Securities specifically noted in its report that the CPO industry is currently exhibiting a positive trend of "supply-demand resonance."

On the demand side, AI chip manufacturers led by NVIDIA continue to release GPU products with ever-higher computing power, driving corresponding network bandwidth requirements upward. When building ultra-large-scale AI training clusters, major cloud providers have increasingly stringent requirements for network power consumption and interconnect efficiency, creating clear and urgent demand scenarios for large-scale CPO deployment.

On the supply side, after years of technological accumulation and industry validation, CPO's core optical components, packaging processes, and system integration solutions have gradually matured. Multiple supply chain companies have achieved key technological breakthroughs and possess mass production and delivery capabilities. Supply-side maturity provides a solid industrial foundation for CPO's accelerated penetration.

Three Categories of Supply Chain Companies Worth Watching

Sinolink Securities recommends that investors focus on three categories of core companies in the CPO supply chain:

• CPO core optical component companies: Suppliers of essential components including silicon photonics chips, lasers, and photodetectors — these companies control the "heart" of CPO technology.
• CPO manufacturing companies: Manufacturers with advanced packaging capabilities that can achieve high-precision co-packaging of optical engines and electronic chips, representing a critical link in CPO solution deployment.
• CPO solution companies: Enterprises capable of providing complete CPO solutions from chip to system, helping downstream customers achieve a smooth transition from traditional pluggable optical modules to CPO architectures.

Outlook: CPO May Become the Next Breakout Point in AI Infrastructure

From a broader perspective, the accelerated penetration of CPO technology is an inevitable trend in the ongoing evolution of AI infrastructure. As AI large model parameter scales advance toward the trillion level and GPU counts per cluster continue to increase, bottlenecks in traditional optical interconnect solutions will become ever more pronounced. As the core solution for next-generation optical interconnect technology, CPO's strategic value is being reassessed by both the industry and capital markets.

Industry analysts believe that 2025 to 2026 is likely to be the critical window for CPO technology to transition from small-batch validation to large-scale deployment. Companies in the supply chain that have completed their technology roadmaps and production capacity preparations first will secure a first-mover advantage in this wave of technological transformation, ushering in rapid performance growth.