SoftBank has entered the AI infrastructure race, introducing a sovereign AI Data Center GPU Cloud service designed to keep data processing within Japan. This marks its shift from traditional telecom operations to competing directly with global cloud titans like Amazon, Google, and Microsoft. This significant move is part of its broad “Activate AI for Society” strategy, emphasizing the importance of data sovereignty in today’s digital age. Japanese enterprises, wary of international data protocols, may find this solution particularly appealing.
Interestingly, the service is powered by Nvidia technology, integrating the telecom edge network with central GPU data centers. This combination offers an innovative solution that utilizes SoftBank’s telecom assets, AITRAS edge nodes, and AI computing power, promising users seamless data processing within national boundaries and not requiring dependence on foreign cloud systems.
SoftBank launched a beta version immediately, with full commercial availability planned for October 2026. Initially, it will be for internal use before expanding to external clients. This strategic approach allows SoftBank to fine-tune its offering based on internal feedback, potentially leading to a robust service once it opens to the market.
Central to the service is SoftBank’s Infrinia AI Cloud OS. This proprietary software stack combines AI computing capabilities with necessary layers to efficiently manage AI workloads. Customers benefit from options like Kubernetes as a Service (KaaS) for managing containerized workloads, and Inference as a Service (Inf-aaS) which provides model inference through APIs. This setup promises reduced operational costs and provides clients with a comprehensive, ready-to-use AI platform.
The relationship with Nvidia is pivotal, with the cloud utilizing Nvidia’s GB200 NVL72 systems based on Grace Hopper architecture. It also incorporates Nvidia BlueField-3 DPUs for advanced networking tasks, and Spectrum Ethernet switches aid in delivering precise 5G timing protocols. This blend of technology facilitates efficient support for AI and vRAN workloads, enhancing both AI throughput and network efficiency.
SoftBank envisions this AI Data Center GPU Cloud as a core element of its Telco AI Cloud strategy. This vision aims to create next-generation infrastructure by marrying large-scale data center capabilities with multi-access edge computing. AITRAS, their software-defined AI-RAN solution, already in use at Nvidia’s Santa Clara HQ, supports this edge-focused approach by allowing swift processing with minimal latency. The infrastructure, shared between AI and telecom functions, allows SoftBank to achieve efficient utilization, effectively receiving enhanced performance for AI and 5G from the same framework.
Through this move, SoftBank transforms its national telecom network from a mere operational asset into a strategic advantage in AI infrastructure provision. This approach not only targets a market void but also presents a model of how telecom operators can diversify and add value in a rapidly evolving digital landscape. The emphasis on keeping data within Japanese jurisdiction taps into a rising demand among businesses for sovereign processing solutions, setting SoftBank apart from global hyperscalers who lack locally-focused offerings.
