SoftBank Corp. launched AITRAS, an AI-RAN-based convergence solution designed to host AI and RAN (Radio Access Network) workloads on the same NVIDIA-accelerated computing platform. The solution, powered by the NVIDIA GH200 Grace Hopper Superchip, integrates carrier-grade RAN with AI applications like generative AI, enabling high capacity, performance, and quality. SoftBank plans to roll out AITRAS across its own network and offer it to global telecom operators starting in 2026.
AITRAS leverages a unique L1 software layer, developed on NVIDIA’s AI Aerial platform, to optimize signal processing and task timing while enhancing RAN capacity and reducing power consumption. The system includes an orchestrator to dynamically allocate computing resources, improving cost efficiency for AI and RAN operations. The solution also supports enterprise development and deployment of large language models (LLMs) via NVIDIA AI Enterprise software.
SoftBank aims to provide an AITRAS reference kit to telecom operators in 2025, allowing them to test and deploy the solution in their labs. This kit includes seven key components such as the orchestrator, edge AI applications, virtualization platform, and radio units. By integrating AI and RAN on the same infrastructure, telecom operators can enhance network performance while exploring innovative AI use cases.
Key Points:
• AITRAS enables AI and RAN to run on the same NVIDIA GH200 Grace Hopper Superchip platform.
• L1 software ensures high stability and efficiency for carrier-grade RAN.
• Orchestrator optimizes resource allocation to cut costs and power usage.
• Reference kit includes seven components, supporting telecom operators’ testing and deployment.
• Global rollout of AITRAS to begin in 2026, with reference kits available in 2025.
Junichi Miyakawa, President & CEO of SoftBank Corp., stated, “As every industry is redefined through AI, AITRAS will become an essential element in realizing next-generation social infrastructure, contributing to a sustainable society.”
