Advanced workload shifting social demonstration project launched

In recent years, the rapid spread of AI has led to a surge in demand for data centers (DCs) that serve as its computing foundation. This has resulted in serious challenges in Japan, such as the concentration of DCs in urban areas and the associated power supply issues. On the other hand, the increasing curtailment of renewable energy (RE) due to its expanded introduction in regional areas also presents a challenge. If this situation continues, Japan may not be able to meet its domestic AI computing demand, leading to a risk of important data flowing overseas along with computing demand through the use of foreign cloud services, and the risk of critical services supporting the digital society being disrupted during large-scale disasters.
To avoid such complex risks, WLS technology, which dynamically shifts computing processes (workloads) between distributed DCs in Japan, taking into account power supply and demand conditions and computing resource operating status, is essential, rather than concentrating computing processes in specific DCs. This Demonstration aims to implement an advanced WLS mechanism that coordinates communication, power, and computing across a wide area, and to verify its technical, operational, and business feasibility.

August, 2026 –February, 2027

The Demonstration will establish a wide-area network utilizing the Science Information NETwork (SINET) operated by the National Institute of Informatics (NII) and optical fibers for power security (dark fiber) held by TEPCO PowerGrid. Additionally, the configuration of directly connecting some DC sections with an all-optical network will be verified. By connecting academic and industrial DC clusters to this network, an environment capable of collaborative Wide-Area WLS will be realized.
During WLS execution, an orchestrator[3], functioning as the heart of the Virtual Hyperscaler, will play a role in optimizing the shifting method. Specifically, by utilizing the orchestrator, information such as power supply and demand, market information, GPU utilization, battery status, and network conditions will be collected and analyzed to determine and control workload placement. 
To verify the elements necessary for social implementation, multiple sub-projects are planned. These will be implemented in stages, from foundational demonstrations establishing the basics of Wide-Area WLS between universities nationwide to advanced demonstrations verifying the effectiveness of WLS technology, including integrated control with RE and storage batteries, and orchestration.