Elementary, Mr. Watson: Controlled Impedance—When ‘Connecting the Dots’ Stops Working
Beyond Design: How Signals Survive the Hostile PCB Environment
Connect the Dots: What Designers Should Know About Non-conductive Via Fill
TrendForce’s recent research on silicon photonics (SiPh) shows that the rapid growth of AI training and inference workloads is pushing AI data centers toward increased power use, higher rack densities, and larger clusters. As data transfer becomes a major energy drain, CSPs are treating interconnect technologies as equally important as compute hardware. The architecture of interconnects now plays a key strategic role in determining AI factory growth, energy efficiency, and supply chain management.
Consequently, TrendForce predicts that the markets for co-packaged optics (CPO) and near-packaged optics (NPO) will expand significantly, from around US$100 million in 2025 to over $39 billion by 2030.
CSPs accelerate NPO ecosystem development while CPO targets high-power integrated applications
TrendForce notes that as data transmission speeds advance from 100 Gbps per lane to 200 Gbps per lane and eventually toward 400 Gbps per lane, the limitations of conventional copper interconnects, including signal loss, compensation costs, and power consumption, are becoming increasingly apparent.
Bringing optical connectivity closer to switch ASIC, shortening electrical paths, and reducing system power consumption have therefore emerged as key design priorities for next-generation AI data centers. Consequently, three major technology pathways—linear pluggable optics (LPO), NPO, and CPO—are attracting significant industry attention.
From a deployment perspective, NPO remains the preferred near- to mid-term transition solution for many CSPs. Its advantages include shorter electrical transmission distances and lower power consumption, while preserving modularity, serviceability, and flexibility in multi-vendor sourcing. Chinese CSPs such as Alibaba and Tencent have identified NPO as a core medium-term strategy and are promoting related open standards through the Open Data Center Committee (ODCC).
Meta and Microsoft have likewise prioritized NPO development, leveraging the Open Compute Interconnect Multi-Source Agreement (OCI-MSA) to foster an open optical interconnect ecosystem. Amazon has adopted a multi-vendor strategy and is collaborating with STMicroelectronics on NPO-related initiatives.
By contrast, CPO is better suited for long-term applications requiring higher power density, tighter integration, and greater system performance. Within NVIDIA’s ecosystem, some smaller and mid-sized CSPs are more inclined to adopt NVIDIA’s fully integrated CPO-based AI systems, citing advantages in system integration, deployment efficiency, and platform consistency.
However, TrendForce emphasizes that large-scale commercialization of CPO still faces several challenges, including manufacturing yield, serviceability, fiber-connector standardization, and constraints in InP laser supply. Scale-out CPO switches must integrate a large number of optical engines, placing significant pressure on overall system-level yield.
Meanwhile, removable fiber connectivity approaches, including solutions from Corning GlassBridge, Teramount, the Senko alliance, and Intel OCI, continue to evolve in parallel, which reflects a highly fragmented and innovative technology landscape.
Optical infrastructure race begins as AI factory competitiveness depends on connectivity
Although widespread CPO adoption may take longer due to technical complexity, optical communications infrastructure has already emerged as the next battleground in AI infrastructure development. End users are increasingly locking in access to critical resources such as lasers, photodetectors, InP substrates, and optical fiber to secure future AI Factory expansion capabilities.
Since 2024, the supply of InP substrates has remained tight, while lasers and photodetectors have become strategic components aggressively pursued across the industry. AMD, for example, has recently accelerated procurement efforts for external laser solutions and is reportedly negotiating major purchase agreements for high-power CW laser chips to ensure future capacity availability and to avoid supply-chain dependence tied to NVIDIA's ecosystem and other leading CSPs.
Meanwhile, fiber-optic leader Corning has received investments, support for capacity expansion, and long-term procurement commitments from Meta, NVIDIA, and Amazon. These developments highlight how major CSPs increasingly view optical fiber as a strategic asset within AI infrastructure.
TrendForce forecasts that the CPO/NPO market will surpass $39 billion by 2030, with growth accelerating sharply between 2028 and 2029 as scale-up architectures begin incorporating optical interconnect technologies. Meanwhile, the pluggable optical transceiver market is expected to remain sizeable, maintaining a market value of nearly $26 billion in 2030.
This suggests that the future of optical interconnects will not be dominated by a single technology. Instead, multiple architectures will coexist, with adoption determined by factors such as power efficiency, transmission distance, cost, technology maturity, and supply chain control requirements across different application scenarios.
