NTT DOCOMO BUSINESS examines how optical networking and distributed infrastructure could reshape AI compute architectures through its “GPU over APN” concept. The presentation explores how IOWN’s All-Photonics Network (APN) can connect geographically distributed data centres with ultra-low latency and high-bandwidth optical connectivity, enabling AI training, inference and high-speed data transfer across multiple sites with performance close to single-site deployments. The session also highlights proof-of-concept work involving distributed NVIDIA GPU clusters, long-distance parallel inference and 800Gbps RDMA data transfer, alongside the broader challenge of scaling AI infrastructure sustainably as GPU demand, power consumption and data centre constraints continue to grow.
