Nvidia unveils OOI co-packaged optics at SEMICON West 2026

At SEMICON West 2026, Nvidia presented OOI (Optics-on-Interposer), its next-generation co-packaged optics (CPO) approach. According to a report from Eastmoney on 31 October 2026, OOI departs from the current CPO path by moving the optical engine directly onto the silicon interposer that carries the GPU, rather than routing it through an organic package substrate. In doing so, it brings the photoelectric conversion step closer to the chip core. Nikola Nedovic, a principal research scientist at Nvidia Research, explained that the approach replaces the conventional organic substrate with a larger interposer, and Nvidia already has prototype samples in hand.

The report also places Nedovic at the center of Nvidia's high-speed interconnect and silicon photonics I/O research, where he works on bandwidth and power constraints across multi-GPU and multi-chiplet designs and supports interconnect scaling for superchip clusters such as GH200. He is described as a key contributor to Nvidia's push into co-packaged optics and silicon photonics.

The coverage contrasts OOI with what is in volume production today. Existing CPO designs already pull the optical engine away from pluggable modules and place it next to the switch silicon on an organic substrate, cutting electrical signal loss from roughly 22dB in traditional setups to about 4dB. Even so, signals must still cross two media, the silicon interposer and the organic substrate, before electrical-to-optical conversion happens. OOI instead uses short, fine-pitch wiring to place the optical engine on the same interposer as the compute chips, so signals no longer detour through the organic substrate, improving transmission efficiency and offering a new interconnect direction for next-generation hyperscale AI clusters. As of the latest disclosure, the work stands at the prototype-sample stage, with no mass-production or deployment timeline given.