11 Mar 2027
12:00 - 13:00
Expo Theater I
The investment in AI/ML infrastructure, which for 2027 is projected to exceed $1 trillion and, by some estimates, would reach $1.5 trillion, has led to an equally explosive growth in data center optics and to a new photonics technology ecosystem optimized for AI/ML interconnects. As discussed in the previous 3 OFC IEEE panels, this new AI/ML photonics ecosystem has focused on lowering the power (pJ/bit), latency, and cost ($ per Gb/s) of optical interconnects, initially through LPO/LRO, then CPO and NPO, and more recently also by the transition beyond 200G PAM4 to slow-and-wide NRZ optical interconnects (e.g., OCI WDM MSA) that enable scale-up clusters to extend beyond copper interconnects. Complementary to the evolution of optical interconnects, the use of OCS (Optical Circuit Switching) systems has also been proposed as another significant advancement for ML/AI datacenter networks. Most notably, Google has advocated the value of OCS and reported using OCS extensively for multiple data center networking generations (initially in their “Jupiter” front-end, and then in their TPU scale-up clusters). The OFC 2027 IEEE panel will invite senior leaders from hyperscalers and ML/AI infrastructure providers to review the system and network implications of OCS in the new AI/ML photonics ecosystem. It will aim to debate, beyond hype, the main benefits and requirements for the different use cases in scale-up, scale-out, and scale-across. In addition to the OCS design requirements, the panel, building on previous years’ panels, will also discuss the broader implications in the AI/ML optical interconnect ecosystem, such as the importance of new standards, MSAs, and supply chain maturity.
Presented By:
