From the perspective of the CPO market, an even bigger opportunity than GPU-CPO development is Ethernet switch CPOs. There are thousands of Ethernet switches in data centers, and their share of power consumption is significant, and the transition from 400-800G port speeds to 1.6-3.2T requires a new architecture. Nvidia and Broadcom have already developed 1st generation switch CPOs, which utilize SiPh photonics and have demonstrated the concept’s functionality. They still require an external laser, but real added value will only come from a solution containing an optical engine, which is a truly significant opportunity for Nokia. Currently, the added value Nokia receives from its data center switches, which utilize Broadcom’s ASICs (Jericho), is relatively small and, due to the business model, clearly smaller than, for example, Arista (even though Arista is also very dependent on Broadcom). When Nokia integrates its own optical engine into Ethernet switches, the competitiveness of the switches and the added value Nokia receives will increase significantly. And the development is not limited to this yet; a similar electrical ASIC-SerDes unit is also in routers, and in the future, Nokia can replace pluggable optics with an integrated optical engine in routers as well. Pluggable sales will decrease, but product competitiveness and margins will improve. Perhaps this is “That which must not yet be spoken of” because it would be “reckless.” Of course, Nokia’s competitors are also preparing for the CPO leap, but they lack their own InP-PIC Fab.
Here is a related GTP5.2 summary: It is not an oversimplification to say that the ASIC–SerDes electrical unit is a key part of data transmission at the GPU, switch, and router level. It is a common interface where electrical computation meets the physical transmission path, whether it is copper or optics. CPO and optical I/O do not eliminate this unit, but rather move the optics physically closer to it and make its role even more critical.