NVIDIA and SPIL Target 15-Fold Optical-Engine Expansion in Taiwan

NVIDIA and Siliconware Precision Industries Co., Ltd. (SPIL) plan to expand optical-engine manufacturing capacity in Taiwan 15-fold by mid-2027, adding capacity for equipment that connects large AI computing systems.

The partners intend to double the expanded capacity again by the end of 2028. Both stages are manufacturing targets announced Oct. 21, with the first increase due in 2027 and a further expansion planned for the following year.

Optical engines move data through fiber connections. In co-packaged optics, they sit beside a switch chip instead of operating in separate plug-in transceivers. The equipment connects the computing systems used for large AI workloads.

NVIDIA’s Spectrum-X Ethernet Photonics switch also uses co-packaged optical engines. Each switch package integrates 32 engines rated at 3.2 terabits per second apiece. Spectrum-X Ethernet Photonics was in full production by June 1.

SPIL provides chip-scale packaging, assembly and testing for the platform. TSMC fabricates the silicon photonics, while Foxconn assembles the switch systems. SPIL’s work sits between photonic-chip fabrication and integration of the finished networking equipment.

Production testing has drawn attention elsewhere in co-packaged optics manufacturing. Advantest, VIAVI and SENKO demonstrated a simulated production test flow for co-packaged optics modules after announcing a collaboration in Sept. 2026. The Taiwan plan adds an optical-engine capacity target alongside that broader work on production processes.

NVIDIA is also working with Cisco on a pluggable optical module for AI data centers that targets up to 12.8 terabits per second. That proposed module uses a different approach to packaging optics than a switch with engines beside its chip.

“Demand is extraordinary. Capacity is the constraint,” NVIDIA founder and CEO Jensen Huang said when the expansion plans were announced Oct. 21.

Beyond optical engines, NVIDIA aims to increase product-substrate capacity 2.5-fold over three years. The substrate target and Taiwan plan cover separate parts of its manufacturing supply chain.

NVIDIA has also committed a $1.1 billion capacity-reservation advance for King Yuan Electronics’ semiconductor testing expansion in the United States. High-volume production at those U.S. facilities is targeted for April 2027, ahead of SPIL’s mid-2027 optical-engine capacity goal.