**SoftBank reveals its proprietary AI gateway 'Cloud Proxy' supporting the '1 person, 100 agents' vision**
The AMW Read
Novelty: SoftBank's Cloud Proxy is a concrete enterprise gateway deployment from a tier-1 telco, affirming the agent-infrastructure trend with specific architectural choices. Significance: segment-level impact for enterprise AI infrastructure as it demonstrates production-scale multi-agent orchestra
**SoftBank reveals its proprietary AI gateway 'Cloud Proxy' supporting the '1 person, 100 agents' vision**
SoftBank has announced the internal development and deployment of "Cloud Proxy," a proprietary AI gateway infrastructure designed to support the company's ambitious "1 person, 100 agents" vision. The system, conceived in 2023 and now serving over 240 systems and 20,000 users, acts as a centralized intermediary between internal teams and multiple large language model (LLM) APIs including Azure OpenAI Service, Gemini, and the domestic Sarashina model. It uses unique Internal Keys per tenant, replacing direct API-key exposure to improve security, observability, and token-cost tracking across the enterprise.
**Why it matters.** This is a telling case study of the enterprise agent-infrastructure bottleneck — the "five walls" SoftBank explicitly names: latency, configuration time, security, governance, and scalability. The "1 person, 100 agents" vision is the extreme version of the multi-agent enterprise pattern described in our industry substrate, where the rate-limiting step shifts from model capability to the orchestration/secirity layer. SoftBank's choice to build a proprietary gateway rather than rely on a third-party vendor reflects a recurring pattern among large incumbents (telcos, financial services) that view agentic traffic as strategically sensitive infrastructure, not a commodity.
**Grounded expert take.** The article's detail on the Cloud Proxy's architecture — per-tenant Internal Keys rather than shared API keys, centralized token audit, sub-5-minute scale-out to handle sudden traffic spikes — mirrors the "context-engineering moat" and "hyperscaler distribution" patterns in our framework. SoftBank is effectively building the enterprise equivalent of an API management gateway for the agent era. The 70% reduction in AI application onboarding time and the ability to add new model endpoints in one week (versus months previously) are concrete operational metrics. The key open debate here is whether such proprietary gateways become standard operating procedure for large enterprises, or whether third-party platforms (e.g., Azure API Management, Kong, or AI-specialized middleware) eventually capture this workload.
