🗾ITmedia AI+ (日本)•Stalecollected in 2h
SoftBank Launches Sovereign LLM Service

💡Japan's sovereign LLM service secures enterprise AI with domestic DC control
⚡ 30-Second TL;DR
What Changed
Enterprise service using Sarashina LLM launches in June
Why It Matters
Offers Japanese firms compliant LLM access without foreign data risks. Strengthens enterprise AI adoption in regulated sectors. Highlights sovereign AI trend in Asia.
What To Do Next
Sign up for SoftBank Sarashina beta to test sovereign LLM for sensitive data apps.
Who should care:Enterprise & Security Teams
🧠 Deep Insight
AI-generated analysis for this event.
🔑 Enhanced Key Takeaways
- •The Sarashina LLM is specifically optimized for the Japanese language and business context, leveraging SoftBank's proprietary datasets to outperform general-purpose models in domestic regulatory compliance and nuanced corporate communication.
- •The partnership with Oracle utilizes the 'Oracle Cloud Infrastructure (OCI) Dedicated Region' architecture, allowing SoftBank to maintain physical control over the hardware while benefiting from cloud-native AI orchestration tools.
- •SoftBank is positioning this service as a key component of its 'AI-RAN' (Radio Access Network) strategy, aiming to integrate LLM processing directly into 5G/6G infrastructure to reduce latency for enterprise edge computing.
📊 Competitor Analysis▸ Show
| Feature | SoftBank Sarashina | NTT Data (tsuzumi) | NEC (cotomi) |
|---|---|---|---|
| Focus | Sovereign Cloud/Oracle | Lightweight/On-prem | Enterprise/Public Sector |
| Deployment | Dedicated OCI Region | Private Cloud/Edge | Hybrid/On-prem |
| Language | Japanese-optimized | Japanese-optimized | Japanese-optimized |
| Pricing | Enterprise Subscription | Custom/Project-based | Custom/Project-based |
🛠️ Technical Deep Dive
- •Model Architecture: Based on a transformer-based decoder-only architecture, fine-tuned specifically for high-context Japanese business terminology.
- •Data Sovereignty: Implemented via OCI Dedicated Region, ensuring data residency within Japan and physical isolation from public cloud multi-tenant environments.
- •Integration Layer: Utilizes RAG (Retrieval-Augmented Generation) pipelines that connect directly to internal enterprise databases (ERP/CRM) via secure, encrypted APIs.
- •Inference Optimization: Employs model quantization techniques to balance high-accuracy performance with the compute constraints of domestic data center hardware.
🔮 Future ImplicationsAI analysis grounded in cited sources
SoftBank will capture significant market share in the Japanese public sector by 2027.
The combination of strict data sovereignty and Oracle's established government-grade security certifications provides a unique competitive moat for handling sensitive administrative data.
The Sarashina model will be integrated into SoftBank's mobile network operations.
SoftBank's strategic pivot toward AI-RAN suggests that the LLM will eventually automate network optimization and predictive maintenance tasks.
⏳ Timeline
2023-11
SoftBank announces the development of a proprietary Japanese LLM.
2024-05
SoftBank and Oracle Japan announce a strategic partnership to build sovereign AI infrastructure.
2025-02
SoftBank begins internal testing of the Sarashina model for corporate administrative tasks.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: ITmedia AI+ (日本) ↗