SoftBank Unveils Telco AI Cloud Vision for Distributed AI Infrastructure

SoftBank Group Corp. announced in a press release its new Telco AI Cloud vision, aimed at creating distributed AI infrastructure by combining its telecommunications network with large-scale GPU cloud data centers, AI-RAN-based edge computing, and the Infrinia AI Cloud OS software stack. The initiative seeks to optimize AI processing from training to inference while maintaining low latency and high reliability.

The Telco AI Cloud integrates three main components: GPU cloud data centers for training, an AI-RAN-based MEC platform for inference, and the Infrinia AI Cloud OS for centralized management. SoftBank's AITRAS Orchestrator dynamically allocates computing resources across AI and radio access network domains based on resource availability, application needs, and power consumption.

As part of the rollout, SoftBank announced collaborations with Ericsson and Mitsubishi Heavy Industries. A proof-of-concept with Ericsson demonstrated low-latency, high-reliability Physical AI networks, while work with Mitsubishi Heavy Industries aims to expand edge data center deployment for industrial AI applications. The company also open-sourced the Dynamic Scoring Framework, a core AITRAS Orchestrator function, to support wider AI-RAN adoption.

SoftBank is further integrating its orchestration technology with Ericsson’s and Nokia’s AI platforms to enable dynamic resource allocation and external AI workload execution. Participation in the Linux Foundation’s OCUDU initiative will support the development of an open, global AI-RAN ecosystem.