TOKYO — Marking a definitive shift from a traditional telecommunications carrier to an AI-native infrastructure provider, SoftBank Corp. officially launched its Multi-AI Agent Platform on Thursday, March 12, 2026. Built within the company’s proprietary Large Telecom Model (LTM), the platform moves beyond simple data analysis, enabling a hierarchy of autonomous agents to execute complex network operations ranging from real-time traffic optimization to critical fault response.
The announcement, delivered by CEO Junichi Miyakawa, follows a series of high-stakes demonstrations at Mobile World Congress 2026 and signals SoftBank’s ambition to become the “central nervous system” of a digital-first society.
The Architecture of Autonomy: How Multi-Agent Systems Work
The platform is designed to eliminate the “siloed” nature of network management. Rather than requiring human operators to bridge the gap between detection and action, SoftBank’s multi-agent system creates a continuous, autonomous workflow:
- Specialized Intelligence: Individual AI agents are optimized for specific roles—such as quality improvement, equipment maintenance, and traffic analysis.
- Mutual Coordination: Using a collaborative logic, one agent detects an anomaly (e.g., a surge in low-latency gaming traffic), analyzes the requirement, and hands over a decision to a “routing agent” to execute a configuration change instantly.
- Eliminating “Personalization”: By automating these high-level decision paths, SoftBank aims to reduce reliance on the “tribal knowledge” of skilled personnel, ensuring consistent, 24/7 network stability.
The “LTM” Advantage: A Foundation for Physical AI
At the core of this transition is the Large Telecom Model (LTM). Unlike generalized LLMs, the LTM is a purpose-built foundation model trained on massive datasets of telecom-specific KPIs, configurations, and domain expertise.
“We are no longer just moving raw, uninterpreted data,” SoftBank stated during its strategy unveil. “We are providing meaning. By embedding intelligence from the core to the edge, we enable robots and autonomous systems to perform complex behaviors powered by the network intelligence that surrounds them.”
Strategic Verification: Autonomous Routing
Concurrent with the platform launch, SoftBank verified its “Autonomous Thinking Distributed Core Routing.” In commercial field trials, the system demonstrated a 99.7% traffic control accuracy.
- The Result: For applications like cloud gaming, the AI agents were able to dynamically switch network routes to satisfy strict latency requirements (under 40ms) without any manual intervention.
- The Impact: This technology allows resource-constrained devices—such as warehouse robots or autonomous vehicles—to “offload” heavy computation to the network’s edge, effectively giving them a “shared brain” via the 5G/6G infrastructure.
The Global Ambition
SoftBank’s pivot toward “Agentic AI” is not limited to its own network. Through partnerships with Northeastern University, NVIDIA, and OpenAI, the company is open-sourcing its Dynamic Scoring Framework (DSF) to accelerate the global adoption of AI-RAN (Radio Access Network) technology.
As the 2026 regional conflict continues to strain global connectivity, SoftBank’s vision of a self-healing, self-optimizing “AI Cloud” represents a significant leap toward a world where the network doesn’t just connect—it thinks.
Masayoshi Son, Chairman and CEO, SoftBank by FT on Wikimedia
