Nvidia and SPIL Plan Major Co-Packaged Optics Capacity Expansion for AI Data Centers
MissedBlock Desk · · 3 min read
Updated
Nvidia and Siliconware Precision Industries (SPIL) are planning a significant expansion of co-packaged optics (CPO) optical-engine capacity, aiming for a 15-fold increase by mid-2027 and a further doubling to 30 times by the end of 2028. This strategic move addresses the escalating demand for advanced optical networking solutions in AI data centers, driven by the limitations of current copper interconnect technology.
According to Nvidia CEO Jensen Huang, the demand for optical components in AI data centers could increase by as much as 15 times. This surge is attributed to the physical constraints of copper interconnects, which are struggling to keep pace with the power consumption, signal degradation, and bandwidth density requirements of modern AI workloads. Optical interconnects offer a way to bypass many of these limitations.
Understanding Co-Packaged Optics
Co-packaged optics (CPO) technology integrates optical engines directly alongside silicon switch chips. This integration reduces the electrical path between the chip and the fiber, leading to lower latency and improved power efficiency. Nvidia is introducing its Spectrum-6 CPO switches, intended for use in its Vera Rubin AI factories, promising higher bandwidth density and reduced energy consumption compared to existing networking equipment.
The bottleneck in CPO deployment has historically been the ability to produce optical engines at scale, rather than the demand for them. The planned expansion by Nvidia and SPIL indicates a concerted effort to overcome this manufacturing hurdle, viewed as part of a larger infrastructure buildout to support the exponential growth in AI.
Ecosystem Investments Support Expansion
In parallel with these CPO expansion plans, Nvidia has committed $1.1 billion to advance US semiconductor testing capacity through a partnership with KYEC, with production targeted to begin in April 2027. Additionally, Amkor plans to increase its interposer wafer production capacity by over 300% by the end of 2028. These related investments suggest a broader ecosystem-wide effort to scale up manufacturing capabilities for advanced AI infrastructure components.
The economics and engineering for large-scale CPO deployment have recently aligned, making practical implementation feasible. The accelerated timeline for this shift is largely due to the demands of AI training clusters. The planned capacity increases by Nvidia’s partners, such as the more-than-tripling of capacity signaled by Amkor, indicate strong confidence in continued demand growth for these critical components.
Why This Matters
This planned expansion by Nvidia and SPIL is critical because it directly addresses the limitations of current copper interconnect technology, which is straining under the demands of modern AI workloads. Optical interconnects, and specifically co-packaged optics (CPO), offer a solution to the power consumption, signal degradation, and bandwidth density challenges posed by copper. The significant capacity increase signals a strategic shift towards optical networking as essential infrastructure for the continued exponential growth of AI data centers.
Broader Context
These developments are occurring within the context of a significant buildout of AI infrastructure. Nvidia’s investment of $1.1 billion in US semiconductor testing capacity with KYEC, and Amkor’s planned over 300% increase in interposer wafer production capacity by 2028, underscore a broader industry-wide effort to scale manufacturing for advanced AI components. The alignment of economics and engineering for CPO deployment, accelerated by AI training clusters, makes this expansion timely and strategically important for meeting future AI demands.