← Back to feed

SoftBank Corp, the telecom arm of SoftBank Group, announced its “AI Data Center GPU Cloud” — a sovereign AI infrastructure service positioned to compete directly with AWS, Azure, and Google Cloud for Japanese enterprise workloads. The differentiating claim: data stays inside Japan, and the GPU compute ships bundled with SoftBank’s 5G network.

Architecture

The service runs on Infrinia AI Cloud OS, a proprietary software stack delivering:

• Kubernetes-as-a-Service (KaaS) for container orchestration
• Inference-as-a-Service (Inf-aaS) for large language model deployment
• AITRAS edge nodes that physically connect the 5G network to central GPU data centers

SoftBank CEO Junichi Miyakawa framed the goal as providing “integrated computing infrastructure and software that can be securely used within Japan as a neocloud provider.” Target launch: October 2026.

SoftBank has not disclosed GPU count, data center capacity, or total capital commitment.

The Telecom Moat Claim

The distinctive play is network bundling. SoftBank is not selling generic GPU compute — it is marketing its 5G infrastructure as a free inclusion on shared hardware, internally framed as “5G for free.” The proposition: if edge compute and networking run on the same AITRAS nodes, SoftBank can deliver latency and integration advantages that a pure-cloud provider cannot match.

That claim will face scrutiny. Most enterprise AI workloads are not latency-sensitive in ways that distinguish a well-peered cloud region from collocated edge nodes. The advantage is real in narrow use cases — real-time inference at the network edge — and marginal in the majority of enterprise workflows.

Japan’s Sovereign AI Race

SoftBank’s domestic rival NTT Data has already moved in the same direction. NTT unveiled its sovereign AI cloud at MWC in March 2026, integrating a Japanese-language foundation model, GPU infrastructure, and agentic services into a unified offering for government and enterprise. SoftBank’s entry means Japan’s two largest telcos are now directly competing for the same sovereign AI contract pool.

The broader context is a global pattern. The EU’s May 27 Tech Sovereignty Package would bar AWS, Azure, and Google Cloud from government health and finance data. France’s AION Consortium placed a $10B bid for an EU AI Gigafactory. Canada committed $9B to a TELUS-anchored sovereign AI cluster. Japan is following the same trajectory, with two domestic operators now positioned to absorb workloads that would otherwise flow to US hyperscalers.

What SoftBank Is Not

The announcement is infrastructure, not a model. SoftBank is not competing with DeepSeek or Gemini — it is competing with Azure Japan, AWS Tokyo, and Google Cloud’s asia-northeast1 region for the right to host inference workloads that regulators or enterprise risk officers want onshore. That is a large and defensible market, but it is a different bet than building frontier AI capability.