SoftBank Corp. and Red Hat Develop Solution to
Optimize Power Consumption in AI-RAN Data Centers

Background of Development

SoftBank is advancing the development of AITRAS, an integrated solution that enables AI and RAN (Radio Access Network) to operate on the same platform, based on the AI-RAN concept. In the future, SoftBank aims to deploy AITRAS data centers equipped with GPU (Graphics Processing Unit) servers across Japan. These data centers will facilitate the enhancement of RAN through AI, improve network quality, and enable the provision of various AI applications.

To ensure the stable operation and optimized performance of vRAN and AI applications within AITRAS data centers, a significant amount of power is required. There is a growing momentum to promote renewable energy adoption to reduce greenhouse gas emissions from fossil fuels, as well as an increasing focus on environmental assessment through carbon intensity (the ratio of CO2 emissions from power generation). At the same time, as most data centers in Japan are concentrated in major metropolitan areas, there is an increasing need to strengthen resilience and geographically distribute power loads more efficiently.

In response to these energy challenges, SoftBank and Red Hat developed a power optimization solution by leveraging their expertise and technologies as part of their joint research and development initiative^*3 to advance AI-RAN, which began in November 2024.

Overview of the Power Optimization Solution

Red Hat derives its power monitoring capabilities from the Kepler^*4 project, delivered as a supported capability in Red Hat OpenShift. Kepler operates in Kubernetes^*5 environments and collects power consumption data (metrics) from servers in each cluster within a data center, as well as from individual applications. This power-related data is then made available to AITRAS Orchestrator and other software. Additionally, Kepler provides estimated power consumption data based on GPU utilization, including virtualized GPU resources partitioned via MIG (Multi-Instance GPU) technology.

SoftBank has added power optimization functionality to the AITRAS Orchestrator by implementing the Kepler software into AITRAS's virtualization infrastructure, which is based on Red Hat OpenShift. As a result, in addition to traditional resource allocation parameters such as data center location, application priority, and GPU allocation size, the AITRAS Orchestrator can now also factor in real-time power consumption data when dynamically assigning AI applications to the most optimal resources.

For example, the AITRAS Orchestrator can optimize resource allocation in the following ways:

• In multi-cluster environments, it can monitor power consumption across clusters and assign applications to clusters with lower power usage.
• Power consumption limits can be set for each cluster, ensuring applications are deployed without exceeding the predefined threshold.
• Renewable energy availability and carbon intensity metrics can be set for each cluster, allowing applications to be deployed based on environmental impact considerations.

This solution is scheduled to be showcased at Red Hat's booth at MWC Barcelona 2025.