On July 6, 2026, NTT Docomo Business launched the "GPU over APN Testbed," a wide-area distributed pilot environment connecting GPU resources across 8 national locations via IOWN APN. This platform integrates hardware located in Sapporo, Kanazawa, Fukuoka, Osaka, and 4 sites in the Tokyo metropolitan area, allowing them to operate as a unified resource. Users can verify the efficacy of distributed AI training, inference, and large-capacity data transfers across sites using high-speed, low-latency 100Gbps communications.

Traditional data center configurations have faced limitations regarding resource acquisition for large-scale models, power supply constraints, and capacity caps at individual centers. IOWN APN addresses these issues by extending optical technology to the network edge, providing high-bandwidth and energy-efficient transmission. This pilot seeks to build GPU clusters that remain performant regardless of geographical distance.

Technically, the system supports virtual machines and Kubernetes, enabling multi-tenant utilization. Specifically, the adoption of RDMA—a technology that allows direct memory transfer without significant CPU overhead—has established a low-latency, high-throughput environment. This provides a platform for validating system configurations and design challenges under real-world operating conditions.

This project serves as a verification phase leading toward the formal launch of a GPU cloud service. NTT Docomo Business plans to accumulate insights through joint trials with customers and partners to drive future service expansion and the societal implementation of distributed AI. These efforts are aligned with the development of the company's AI-native infrastructure, "AIOWN," and its next-generation ICT platform vision.

On July 6, 2026, NTT Docomo Business launched the "GPU over APN Testbed," a wide-area distributed pilot environment connecting GPU resources across 8 national locations via IOWN APN. This platform integrates hardware located in Sapporo, Kanazawa, Fukuoka, Osaka, and 4 sites in the Tokyo metropolitan area, allowing them to operate as a unified resource. Users can verify the efficacy of distributed AI training, inference, and large-capacity data transfers across sites using high-speed, low-latency 100Gbps communications.

Traditional data center configurations have faced limitations regarding resource acquisition for large-scale models, power supply constraints, and capacity caps at individual centers. IOWN APN addresses these issues by extending optical technology to the network edge, providing high-bandwidth and energy-efficient transmission. This pilot seeks to build GPU clusters that remain performant regardless of geographical distance.

Technically, the system supports virtual machines and Kubernetes, enabling multi-tenant utilization. Specifically, the adoption of RDMA—a technology that allows direct memory transfer without significant CPU overhead—has established a low-latency, high-throughput environment. This provides a platform for validating system configurations and design challenges under real-world operating conditions.

This project serves as a verification phase leading toward the formal launch of a GPU cloud service. NTT Docomo Business plans to accumulate insights through joint trials with customers and partners to drive future service expansion and the societal implementation of distributed AI. These efforts are aligned with the development of the company's AI-native infrastructure, "AIOWN," and its next-generation ICT platform vision.