NVIDIA's strategic deployment in the next-generation AI computing architecture is accelerating the transition of optical interconnect technology from the edge of data centers to the core.
On March 11, according to the latest research from TrendForce, as the scale of AI clusters continues to expand, the physical bottleneck of copper cable transmission is becoming increasingly prominent, significantly enhancing the strategic position of optical transmission technology in data centers.
TrendForce predicts that the penetration rate of Co-Packaged Optical Devices (CPO) in AI data center optical communication modules will steadily rise, expected to reach about 35% around 2030.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/4d13cb1d-2c2c-4e97-aad1-680980890f53.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/4d13cb1d-2c2c-4e97-aad1-680980890f53.png"/>
However, Broadcom previously pointed out that while optical transmission is a long-term alternative trend, its core SerDes chip technology is continuously breaking through the physical limits of copper cables. With significant cost and power consumption advantages, Broadcom believes that copper-based solutions will remain the absolute dominant player in the short-distance interconnect market within data centers at least until 2028.
Wallstreetcn mentioned that NVIDIA announced last week an investment of $2 billion each in Lumentum and Coherent, totaling $4 billion, and signed multi-year procurement agreements to prioritize the supply of advanced lasers and optical devices.
This move is seen as an important layout for NVIDIA in key optical components, and also indicates that the future AI computing infrastructure will increasingly rely on optical technology.
<h2>Copper Cables Reach Physical Limits, Accelerating Demand for Optical Transmission</h2>
TrendForce pointed out that when chip interconnect scales expand from single racks to cross-rack deployments, such as a 576-GPU cluster composed of eight NVIDIA NVL72 systems, copper cable interconnects will struggle to meet the required performance and bandwidth demands.
The NVIDIA NVLink 6 communication protocol defines the transmission rate per channel as 400G SerDes, with a single GPU bandwidth limit of up to 3.6 TB/s. At this extreme transmission rate, the electrical signals in copper cables rapidly attenuate with distance, effectively compressing the transmission distance to less than one meter.
In addition, the competitiveness of optical transmission is concentrated in Wavelength Division Multiplexing (WDM) technology. This technology allows multiple wavelength signals to be transmitted simultaneously over a single optical fiber, significantly enhancing transmission density, which is an inherent advantage that copper cable solutions cannot match.
NVIDIA's technology roadmap in CPO and silicon photonics adopts TSMC's COUPE 3D packaging technology, stacking and integrating logic chips with photonic chips, and integrating 200G PAM4 micro-ring modulators (MRM) on silicon photonic chips, enhancing the bandwidth density of the optical engine while controlling size and power consumption.
Meanwhile, several major cloud service providers are collaborating with emerging startups to actively develop new optical interconnect solutions, preparing for the next round of bandwidth demands and laying the groundwork for the large-scale commercialization of CPO technology
<h2>The Commercial Progress of CPO: Phased Advancement from Scale Interconnection to Rack Deployment</h2>
TrendForce expects that the optical interconnection technology based on silicon photonics and CPO will first be used for inter-rack (scale-out) data transmission in NVIDIA's Rubin generation, while planning to integrate it into future intra-rack (scale-up) interconnection architectures.
In the near term, the share of CPO in AI data center optical transceiver modules is expected to be only about 0.5% by 2026.
Looking ahead to a longer cycle, as silicon photonics and CPO packaging technologies continue to mature, TrendForce judges that cross-rack scale-up optical interconnection solutions may appear as early as in the Rubin Ultra or Feynman generation.
By around 2030, the penetration rate of CPO solutions based on silicon photonics is expected to reach about 35%. TrendForce also notes that new optical technologies, including advanced optical interconnection architectures and optical I/O, may emerge during this period.

NVDX

NVDL

AVGO

NVDA

NVDU

With NVIDIA's NVLink 6 pushing single GPU bandwidth to 3.6TB/s, the transmission distance of copper cables has been compressed to less than one meter, accelerating the demand for optical solutions in AI data centers. TrendForce predicts that the penetration rate of Co-Packaged Optics (CPO) will rise from about 0.5% in 2026 to about 35% in 2030, with the earliest possible deployment between racks being realized in NVIDIA's RubinUltra. However, Broadcom believes that copper-based solutions will continue to dominate the short-distance interconnect market at least until 2028 due to their cost and power consumption advantages

- NVIDIA is strategically deploying next-generation AI computing architecture, accelerating the transition of optical interconnect technology from data center edges to the core.  
- TrendForce predicts a steady increase in the penetration rate of Co-Packaged Optics (CPO) in AI data center optical communication modules, potentially reaching around 35% by 2030.  
- However, Broadcom asserts that copper-based solutions will remain predominant in short-distance interconnect markets until at least 2028, despite the growing demand for optical transmission as copper cables hit physical limits.

Wallstreetcn

T-Rex 2X Long NVIDIA Daily Target ETF

GraniteShares 2x Long NVDA Daily ETF

Broadcom

NVIDIA

Direxion Daily NVDA Bull 2X Shares

The demand for optical interconnects in AI data centers is accelerating, with the CPO penetration rate expected to reach 35% by 2030