The primary focus of this year’s OFC was CPO and its impact on the market. There were three new CPO-driven MSAs, widespread demos of ELSFPs, and the concurrent announcement by Nvidia and Lumentum that scale-up CPO is coming in 2028. This show report provides detailed analysis on the MSAs, ELSFPs, and eventually OCSs that will form the core of new architectures. We also examine the updates in the pluggable coherent market as well as the new multi-rail line systems that are in development.
CPO is Imminent
Cignal AI wrote over a year ago that “CPO was inevitable but not imminent”, a phrase that appears to be catching on and was the theme of Cisco’s presentation at the CPO workshop. However, in our opinion, the market has advanced, and CPO is now imminent because of Nvidia’s efforts. As a soon to be released Cignal AI report on CPO and ELSFPs will illustrate, the advent of integrated optics appears upon us, and the market viability of CPO will become clear within a year. Either CPO is going to lift off or the TSMC/Nvidia venture will fail, leaving a massive capital investment impact crater. An outcome somewhere in between feels unlikely.
In the concise words of Cisco’s Bill Gartner, Nvidia has adopted a “supplier-push” approach with CPO and it plans to aggressively drive the technology into customer hands. This approach differs from the typical “customer-pull” approach by which a superior tech is developed and the market rushes to adopt it. But the uncertainties of CPO cost/reliability/availability, coupled with the massive tooling investment required to make it viable in volume, mandate the “supplier-push” approach.
Nvidia is trying to kick start that volume. The company picked scale-out because pluggables are the backup plan. Success means Nvidia cements its technical leadership and can roll out this approach for scale up in next gen AI systems like Feynman. The general sentiment among many in the industry is that a “supplier-push” dynamic is reflective of marginal technology being muscled out by a powerful supplier. What if that is not the case and that the technology is better? Then what happens?
We believe that the industry is not sufficiently recognizing the widespread impact of a scenario in which Nvidia’s approach is both technically and economically superior. Granted, we ourselves are not certain that it is, but we do recognize that it is quite possible, while the industry assumes low probability.
These are not conclusions, only the possible outcome of Nvidia’s success (quickly followed by Broadcom):
Clients log in to access full report