SoftBank, in collaboration with NVIDIA, has developed an artificial intelligence radio access network (AI-RAN) capable of running AI inference and 5G workloads concurrently.
Telecommunications operators have shown strong interest in this infrastructure because it promises to optimize network operations and open new revenue streams. A live demonstration in Kanagawa prefecture, Japan, highlighted SoftBank’s AI-RAN powered by NVIDIA technology, delivering high 5G performance while repurposing unused network capacity for AI inference tasks.
Conventional network designs must provision for peak traffic, which typically leaves much of the infrastructure idle—often only about one-third of capacity is used on average. AI-RAN’s shared compute model enables operators to utilize the remaining capacity for AI inference services, turning previously dormant resources into productive assets.
NVIDIA and SoftBank estimate that every dollar invested in AI-RAN infrastructure could generate up to five dollars in AI inference revenue. SoftBank’s internal analysis suggests returns of up to 219% per AI-RAN server added to existing sites when both operating and capital costs are considered.
Ronnie Vasishta, Senior Vice President of Telecom at NVIDIA, said: “Moving from single-purpose to multi-purpose AI-RAN networks can deliver five times the revenue for every dollar of capex. SoftBank’s live field trial is a major milestone toward AI-RAN commercialization, validating feasibility, performance and economics.”
AI-RAN real-world applications
SoftBank demonstrated several practical AI-RAN use cases during the trials.
Using NVIDIA AI Enterprise, SoftBank deployed a variety of edge AI inference applications—examples include remote support for autonomous vehicles, robotics control, and localized automated data retrieval and content generation. These inference workloads ran on SoftBank’s AI-RAN platform without degrading 5G performance.
A notable technical achievement is SoftBank’s fully software-defined 5G radio stack, optimized for NVIDIA’s AI computing environment. This implementation incorporates L1 software enhancements developed by SoftBank using NVIDIA Aerial CUDA-accelerated RAN libraries to tightly integrate RAN functions with AI processing.
Looking forward, SoftBank plans to adopt NVIDIA Aerial RAN Computer-1 systems in its deployment, anticipating roughly a 40% reduction in power consumption compared with traditional 5G infrastructure.
Supply and demand
A core capability of SoftBank’s AI-RAN is dynamic allocation of compute resources to match changing demand while preserving carrier-grade network performance.
SoftBank is building an ecosystem that links AI demand to available capacity. Using NVIDIA AI Enterprise’s serverless APIs alongside SoftBank’s custom orchestrator, the system can redirect external AI inference jobs to AI-RAN servers whenever spare compute resources are available, ensuring efficient utilization of assets.
Ryuji Wakikawa, Vice President and Head of the Research Institute of Advanced Technology at SoftBank, commented: “SoftBank’s ‘AITRAS’ is the first AI-RAN solution born from a five-year collaboration with NVIDIA. It integrates and orchestrates AI and RAN workloads through a SoftBank-developed orchestrator, improving communication efficiency by consolidating dense cells onto a single NVIDIA-accelerated GPU server.
“We believe this AI-driven innovation, AITRAS, will enable new telecom business models and play a pivotal role in mobile operator transformation.”
(Photo by Igor Omilaev)
See also: Elisa and Nokia conduct Europe’s first 100G PON trial
Interested in AI and big data topics from industry experts? The AI & Big Data Expo takes place in Amsterdam, California and London, co-located with events such as the Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.
Discover additional enterprise technology events and webinars presented by TechForge through their events listings.