Nvidia has introduced a prototype of its next-generation co-packaged optics technology, called Optics-on-Interposer, or OOI, at SEMICON West 2026. The new design points to how future AI data centers could handle massive increases in bandwidth demand while keeping power consumption under better control.
As artificial intelligence models continue to grow in size and complexity, the systems that train and run them need faster ways to move data between chips, servers, and networking hardware. Traditional electrical interconnects are becoming a major challenge because they consume more power and face limitations as bandwidth requirements rise. Nvidia’s OOI prototype is designed to address that problem by bringing optical communication closer to the compute chips themselves.
In conventional packaging, the optical engine is typically placed on an organic package substrate. Nvidia’s new approach moves that optical engine onto the silicon interposer, the same advanced platform that carries the compute chips. This shift is important because it allows optical components and high-performance processors to sit much closer together, reducing the distance data must travel and improving overall efficiency.
The main advantage of Optics-on-Interposer is higher bandwidth density. By integrating optics more tightly into the chip package, Nvidia can create interconnects capable of moving larger amounts of data in less space. This is especially important for future AI accelerators, where performance depends not only on raw compute power but also on how quickly information can flow between different parts of the system.
Power efficiency is another major focus. AI infrastructure already requires enormous amounts of electricity, and interconnects play a growing role in total system power draw. By using optical signaling closer to the processor, Nvidia aims to reduce the energy needed to transfer data across high-speed connections. Lower power consumption could help future AI clusters scale more effectively without creating unsustainable energy demands.
The timing of the OOI prototype is also notable. Nvidia is positioning this technology for AI systems expected after 2028, suggesting that co-packaged optics will become increasingly important as next-generation data centers move beyond today’s networking limits. Future AI workloads will likely require much denser and faster communication between GPUs, CPUs, memory, and switching hardware, making advanced optical interconnects a key part of system design.
Nvidia’s Optics-on-Interposer prototype highlights a broader shift in semiconductor packaging. Instead of treating compute, memory, and connectivity as separate pieces, the industry is moving toward highly integrated packages that combine multiple technologies in one advanced structure. Silicon interposers are already used to connect high-performance chips and memory, and adding optical engines directly into that environment could unlock a new level of performance for AI computing.
While this is still a prototype, the announcement gives a clear look at Nvidia’s long-term strategy for AI infrastructure. The company is not only focusing on faster processors, but also on the high-speed communication systems needed to keep those processors fully utilized. As AI models become more demanding, technologies like OOI could play a crucial role in reducing bottlenecks and improving the efficiency of large-scale AI platforms.
Nvidia’s OOI co-packaged optics technology may become a foundation for the next wave of AI data center interconnects, offering a path toward greater bandwidth, lower power use, and more scalable system architectures for the post-2028 era.
