Recently, NVIDIA() announced that its Spectrum-X Ethernet Photonics CPO switch chip has entered full-scale mass production. The first shipments have been delivered to CoreWeave(), Lambda Labs, and Oracle(), making it the world’s first 200G/lane CPO switch system to achieve scaled commercial deployment.
The significance of this milestone goes well beyond a single product launch. It marks the moment when CPO (Co-Packaged Optics) formally left the laboratory: what had been a long-term industry roadmap has become a real purchasing option for AI data centers.
Almost at the same time, SK hynix(**), together with leading academic institutions worldwide, published a CPO technology roadmap paper in **Nature Electronics, pushing the optical interconnect battlefield from the network further toward the memory interface.
One company focuses on compute, the other on storage—yet both have independently placed their bets on the same direction: CPO. That is no coincidence.
Also worth watching is the flurry of recent activity across the CPO landscape: disclosed a CPO roadmap for its MI500 accelerator; TSMC(** **) is accelerating the ramp of PIC capacity; Samsung released a complete plan for a 300mm silicon photonics foundry platform; GlobalFoundries entered the supply chain as a pure-play silicon photonics foundry; and Intel continues to advance compute-native optical I/O. Top chip and foundry companies around the world have all entered the field. A technology race around next-generation compute interconnects is now fully underway.
