Deep|Coherent Lite Reshapes AI Optical Interconnects

Key Takeaways

The industry’s prevailing narrative of Coherent Lite still largely centers on DCI/LR/campus interconnect: the 2–20 km reach segment, where a lower-cost, lower-power coherent solution replaces the conventional IMDD, which is increasingly strained in mid-reach transmission beyond 400G/lane. This narrative is not wrong, but it captures only Coherent Lite’s “reach dimension” and misses the “system-architecture dimension,” where the real volume will come from. From an AI infrastructure standpoint, Coherent Lite’s most important use case — and the one most likely to scale — is not 2–20 km DCI or LR, but rather pairing with OCS in Google TPU / AI clusters.

OCS improves network-topology reconfigurability and expands the Scale-Up and Scale-Out interconnect scale; however, the cost is that each OCS node the optical link passes through adds insertion loss. OCS is fundamentally a passive device, introducing roughly 1.5–3 dB of additional loss. For conventional IMDD FR/DR links, the link budget is typically only a few dB; after subtracting OCS’s ~3 dB of loss and additional fiber/ connector losses, IMDD has almost no remaining link margin in the 2.4T/3.2T generations. In other words, the more OCS becomes the core component of AI cluster network expansion, the more exposed IMDD’s link budget bottleneck becomes.