Co-Packaged Optics Step Into the Spotlight

Optical interconnects in the data center took center stage at NVIDIA's GTC 2025 conference in March when Jensen Huang announced two network switches incorporating co-packaged optics (CPO), which combine photonics and electronics in a single package for higher performance and efficiency.

The announcements validated what Ayar Labs has been saying for years: CPO is not a future technology — it is the essential infrastructure for the AI era. As AI clusters scale to thousands and eventually hundreds of thousands of GPUs, the bandwidth demands exceed what copper interconnects can deliver. Co-packaged optics solves this by moving the optical interface inside the package itself.

Why CPO Is Having Its Moment

The scale-out networking problem — connecting racks of servers in a data center — has long relied on pluggable optical modules. But scale-up, which is the communication within an AI training cluster among GPUs and accelerators, requires much higher bandwidth and much lower latency than pluggable optics can provide.

CPO addresses the scale-up challenge by integrating optical I/O directly into the GPU or accelerator package. This eliminates the copper traces between the chip and the optical module — the bottleneck that limits bandwidth density and increases power consumption.

Ayar Labs' TeraPHY™ optical engine delivers over 8 Tbps per chiplet at less than 25 ns end-to-end latency, with 5x-10x higher bandwidth and 4x-8x better power efficiency compared to traditional copper-plus-pluggable configurations.

The Path to Volume Deployment

With $500M in Series E funding closed in early 2026 and partnerships with leading system integrators including Wiwynn, Ayar Labs is accelerating its path from proven technology to volume production. The spotlight is now on CPO — and Ayar Labs intends to be there at scale.