LightSpeed Photonics is launching 400-Gbps near-packaged optical interconnect modules for AI data centers, addressing the interconnect bottleneck between GPU clusters. The technology aims to reduce latency and power consumption in large-scale model training and inference, where electrical interconnect bandwidth and thermal overhead increasingly constrain scaling.
The product fills a gap in the AI infrastructure stack for large-cluster operators and cloud providers managing dense GPU aggregations, positioning optical I/O as a critical efficiency lever for next-generation training farms.