[News] NVIDIA Unveils Next-Gen CPO Prototype With Optics on Silicon Interposer, Targets Higher Bandwidth Density
NVIDIA is advancing its next-generation co-packaged optics (CPO) technology, with prototypes designed to bring optical engines closer to computing chips. According to Wccftech, the company presented its latest optical integration research at SEMICON West 2026 under the theme “Optical Integration for the Next Generation of AI,” including an optics-on-interposer (OOI) prototype and dense wavelength-division multiplexing (DWDM) technology.
The OOI design moves optical engines from the organic package substrate directly onto the silicon interposer that carries computing chips. According to The Elec, this eliminates the organic substrate from the electrical signal path, allowing signals to travel through shorter, fine-pitch interposer connections that could increase bandwidth density and reduce power consumption.
Beyond OOI, NVIDIA has demonstrated DWDM technology designed to increase data transmission capacity over a single optical fiber. According to Wccftech, the demonstration combines eight wavelengths operating at up to 32 Gb/s each, delivering an aggregate bandwidth of 256 Gb/s per fiber.
Both technologies remain in the research stage, with no commercialization timeline announced. NVIDIA said it is evaluating them for potential future products, according to The Elec. At SEMICON West 2026, NVIDIA researcher Nikola Nedovic also noted that developing efficient optical interconnects for future AI systems requires balancing device and circuit design trade-offs while addressing thermal management and packaging challenges.
Looking ahead, NVIDIA plans to extend optical interconnects with its Kyber NVL1152 architecture, which will connect eight racks into a single 1,152-GPU NVLink domain and support the Feynman platform, according to NVIDIA. Wccftech adds that Feynman is expected to arrive in 2028.
NVIDIA’s Existing CPO Products and Manufacturing Ecosystem
While OOI remains under development, NVIDIA has already brought CPO into commercial production. In May 2026, the company announced that its Spectrum-X Ethernet Photonics switches had entered full production, integrating 200Gb/s SerDes with CPO technology for the Vera Rubin platform. CoreWeave, Lambda, and Oracle Cloud Infrastructure are among the early adopters, according to NVIDIA.
NVIDIA has also established a manufacturing ecosystem supporting its CPO products. According to the company, TSMC provides silicon photonics fabrication, while SPIL handles chip-scale packaging, assembly, and testing. TFC packages laser dies into modules and conducts reliability validation, while Foxconn integrates Spectrum-X Photonics switches into complete networking systems.
Other partners include Coherent, Lumentum, and Sumitomo Electric for external laser source modules; Corning, Senko, and Browave for optical fibers and related components; and Fabrinet for cooling and final testing, as noted by The Elec.
Read more
- Integrated Optical-Electrical Testing Platforms Become Key to CPO Mass Production, Reshaping the Test Equipment Supply Chain, Says TrendForce
- Optical Interconnects Become Critical to AI Factory Expansion; CPO/NPO Market Expected to Exceed US$39 Billion by 2030, Says TrendForce
(Photo credit: NVIDIA)
