Japanese companies test distributed data centers for AI and power optimization

Japanese companies, including NTT Docomo Business and KDDI, have launched demonstration projects to develop distributed data center systems that optimize AI processing and power consumption. These initiatives, selected for the Ministry of Internal Affairs and Communications’ “watt-bit coordination” demonstration program, aim to address the growing shortage of GPU resources and electricity caused by the rise of AI.

NTT Docomo Business, alongside Chugoku Electric Power and other partners, is testing the integration of AI workloads with power information. By connecting data centers in locations such as Mitaka, Sapporo, and Hiroshima via high-speed networks, the project seeks to shift AI inference and learning tasks to regions with available power and computing resources. The goal is to establish a “watt-bit control system” by approximately 2028.

Separately, KDDI, Morgenrot, and Tokunet are conducting trials using retail robots to test distributed AI learning and remote inference. Utilizing All-Photonics Network (APN) technology and GPU virtualization, the group is evaluating whether robots can maintain real-time responsiveness when utilizing GPU resources located in distant data centers. This approach aims to prevent the concentration of computing resources in major metropolitan areas by leveraging available power and GPUs in regional areas.