Nvidia has laid out its next-generation co-packaged optics (CPO) roadmap: move the optical engine off the organic substrate and into the silicon interposer, with prototype samples already in place. Cailian Press reported the disclosure on October 31, with Sina Finance and Sohu Tech following up.
CPO puts the optical engine next to the switch chip to shorten electrical runs and cut interconnect power. The current generation co-packages engines with chips on an organic substrate; the next step, summarized by EETimes China as "OOI silicon-interposer optics," integrates the optical engine into the same silicon as compute and switch logic. Shorter optical paths and higher integration free up power for larger clusters.
The cadence fits a technology moving from ramp to iteration. In August, CPO-equipped switches reached full production; Nvidia has said CPO optical engine capacity will rise 15-fold, and it deepened work with tester-packager SPIL in October. Prototypes in hand mean the next generation has entered engineering validation with a clear path to production.
For the supply chain, the switch moves value around. With engines in the interposer, silicon photonics, interposer manufacturing and advanced packaging gain weight, while pluggable module roles get redrawn. Chinese optical module vendors have positions in the current CPO chain; positioning for the next roadmap has begun.
The AI cluster race has pushed interconnect on stage: as GPU counts double, optical power and bandwidth set the ceiling. The pace of CPO iteration is becoming a hard yardstick for AI infrastructure vendors.