SoftBank launched its sovereign AI Data Center GPU Cloud, pairing Nvidia GB200 systems with AITRAS edge nodes to keep compute and data within Japan.
The company positions the offering as a pillar of its broader Activate AI for Society strategy, with an immediate beta followed by October 2026 commercial availability.
Nationwide network tied to central GPU capacity
SoftBank links its nationwide telecom network to central GPU data centers using AITRAS edge nodes, blending edge connectivity with centralized compute instead of treating AI workloads as a separate infrastructure lane.
The edge component runs on AITRAS, which SoftBank describes as a fully software-defined AI-RAN solution, with that AI-RAN deployment currently at Nvidias Santa Clara headquarters.
Infrinia AI Cloud OS and delivery modes for AI workloads
At the core is SoftBanks proprietary software stack, Infrinia AI Cloud OS, which brings together SoftBanks AI computing infrastructure with the software layers needed to run modern AI workloads at scale.
SoftBank says the platform supports workloads from model training through inference and general data processing, delivered via Kubernetes as a Service (KaaS) and Inference as a Service (Inf-aaS). KaaS is aimed at containerized workloads in multi-tenant environments, while Inf-aaS exposes large language model inference through APIs.
Nvidia-focused build: GB200, BlueField-3, and Ethernet timing
SoftBanks hardware stack is built on Nvidia GB200 NVL72 systems based on the Grace Hopper architecture, hosted in Japan-based data centers and run on SoftBanks neocloud business framework.
On the telecom-and-AI networking side, SoftBank uses Nvidia BlueField-3 DPUs to accelerate both vRAN and generative AI workloads, and pairs that with an integrated Nvidia Spectrum Ethernet switch that provides a 5G timing protocol. SoftBank also says the Infrinia AI Cloud OS stack handles configuration from BIOS up through Kubernetes management on the GPU platforms.
Beta now, internal rollout first, then October 2026
A beta version of the AI Data Center GPU Cloud went live the same day the service was announced, but commercial availability is scheduled for October 2026.
After the beta, SoftBank restricts the initial rollout to internal use across SoftBank group companies before opening the service to wider commercial customers, and it frames the bundling between telecom and AI compute as a way to claim c5G for free.d