Nvidia and SPIL plan to expand optical engine capacity in Taiwan
The move would deepen Nvidia's reliance on Taiwan's packaging ecosystem as its co-packaged optics switches scale up for AI data centers
Nvidia and SPIL plan to expand optical engine capacity in Taiwan.
What Nvidia and SPIL are building
The planned expansion centers on optical engines, the components that turn electrical signals into light so data can travel between chips and switches.
SPIL’s role in Nvidia’s supply chain covers chip-scale packaging and testing.
This sits within Nvidia’s broader co-packaged optics (CPO) push. CPO places optical engines directly onto switch silicon rather than relying on separate, pluggable optical transceivers.
TSMC is also part of the picture. It handles silicon photonics fabrication through its COUPE platform, giving Nvidia two major Taiwanese partners covering different stages of production.
The numbers behind Spectrum-X Photonics
The product driving demand is Nvidia’s Spectrum-X Ethernet Photonics switch platform. Nvidia began shipping select Spectrum-X CPO switches in June 2026.
By mid-August 2026, the company declared full production.
In larger configurations such as the SN6800, the platform reaches throughput of up to 409.6 Tb/s.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Nvidia’s CPO approach reportedly delivers a 5x improvement in power efficiency compared to traditional optical modules. It also uses 4x fewer lasers and offers 10x higher mean time between failures.
The design also supports high-density, liquid-cooled builds.
Who is already buying
The Spectrum-X switches support AI factory networking on Nvidia’s Vera Rubin platform. Cloud providers including CoreWeave, Lambda and Oracle Cloud Infrastructure have adopted it.
Nvidia executives have expressed optimism about ramping capacity in the second half of 2026.
Background: the slow goodbye to copper
Copper interconnects have carried data inside servers and between racks for decades. As bandwidth climbs, copper eats more power and struggles to carry signals far without degradation.
Traditional pluggable transceivers bring extra power draw, extra heat and extra failure points. CPO places optical engines directly onto switch silicon to address these issues.
At SEMICON West in October 2026, Nvidia unveiled prototypes of next-generation “optics-on-interposer” (OOI) CPO. OOI aims to push bandwidth density higher and cut power requirements further.
What this means
Taiwan’s role in AI hardware is widening beyond chip fabrication, with advanced packaging and testing becoming a strategic chokepoint. For SPIL, a larger optical engine footprint would tie it more tightly to Nvidia’s Spectrum-X product line. For Nvidia, it adds packaging capacity in a region where it already leans on TSMC.
Packaging optics directly onto switch silicon raises the stakes on manufacturing yield, since a defect can affect a far more expensive assembly than a single pluggable module.
What to watch next: whether the capacity expansion keeps pace with Nvidia’s second-half 2026 ramp, how quickly OOI moves from prototype to production, and whether more cloud providers beyond CoreWeave, Lambda and Oracle Cloud Infrastructure sign on.