Nvidia and SPIL are planning to increase production capacity for optical engines in Taiwan. This initiative aims to support the scaling of Nvidia's co-packaged optics (CPO) switches, which are designed for AI data centers. SPIL provides chip-scale packaging and testing services for these components.
The expansion complements Nvidia's existing partnership with TSMC, which utilizes its COUPE platform for silicon photonics fabrication. The primary product driving this demand is the Spectrum-X Ethernet Photonics switch platform, which began shipping in June 2026.
The Spectrum-X platform, including the SN6800 configuration, supports throughput of up to 409.6 Tb/s. Nvidia reports that its CPO approach offers five times better power efficiency, uses four times fewer lasers, and provides ten times higher mean time between failures compared to traditional modules.
These switches are integrated into the Vera Rubin platform and have been adopted by cloud providers such as CoreWeave, Lambda, and Oracle Cloud Infrastructure. Nvidia executives aim to ramp up capacity during the second half of 2026.
At the SEMICON West event in October 2026, Nvidia introduced prototypes of its next-generation optics-on-interposer (OOI) technology. This development is intended to further increase bandwidth density and reduce power consumption in future hardware.
