SoftBank and Nokia Plan a 2027 Commercial AI-RAN Field Trial in Japan, On Shared Infrastructure

SoftBank and Nokia have agreed to run a commercial AI-RAN field trial in Japan in 2027, on a live network under real commercial traffic. The trial takes place outside a lab setting. The test will see whether radio and AI workloads can run on the same shared computing infrastructure, using Nokia’s AI-native anyRAN Cloud RAN software alongside SoftBank’s AITRAS platform, with Nvidia’s Aerial RAN Computer, Grace CPUs, RTX PRO 4500 GPUs and ConnectX-8 cards underneath.

The AITRAS Orchestrator is the piece SoftBank is building. It monitors demand for computing resources used for AI processing and RAN control in real time, then allocates resources dynamically based on availability, application requirements and projected power consumption. The trial will assess whether AITRAS can serve as a common platform for hosting RAN software and AI applications from multiple vendors, without being tied to a specific RAN implementation.

The underlying question is resource utilisation. Telecom networks are being asked to carry more data traffic while also providing low-latency, real-time AI processing as generative AI, AI agents and physical AI expand. If the same hardware that handles network traffic can also run AI inference, the operator gets more out of each site. That is the economic case, and it is why Nokia’s CTO-level language frames the trial as a step toward AI-native network architectures.

Nokia’s framing is forward-looking and should be read as such. The company says the trial will contribute to the wider shift toward AI-native architectures by exploring how operators can introduce capabilities faster and support new AI applications at the network edge. The trial itself is the validation step, not the product. Future use cases mentioned include physical AI, immersive digital experiences, robotics and autonomous systems that need close interaction between AI and connectivity.

SoftBank’s Ryuji Wakikawa, senior vice president and CTO, said the field trial marks a significant milestone in moving AI-RAN from concept toward commercialisation, and that SoftBank will keep validating AI-RAN toward commercialisation for networks that deliver both connectivity and AI services. Nokia added that the agreement builds on an ongoing collaboration supported by Nvidia’s accelerated computing node.

The hardware layer is Nvidia’s Aerial RAN Computer, built around RTX PRO 4500 GPUs and ConnectX-8 cards. Nvidia has positioned AI-RAN as a way to turn radio sites into edge inference nodes. GSMA Intelligence puts annual operator capital expenditure on radio access networks in the tens of billions of dollars globally, a figure that matters for any 2027 efficiency claim.

Supplier’s note: We stock Nvidia RTX PRO and ConnectX networking cards for AI-RAN and edge inference builds, and we can quote the full stack for labs and pilot sites running Cloud RAN with GPU acceleration. Lead time on RTX PRO 4500 and ConnectX-8 from our Shenzhen stock is six to eight weeks, and we hold buffer stock for customers running 2027 pilot programmes. For operators or integrators planning an AI-RAN trial, send us your target site count and workload mix and we will come back with a bill of materials and a delivery schedule that matches your trial timeline. We have supplied three AI-RAN pilot builds in the last eighteen months, and the GPU lead time is the item that most often sets the critical path, and we flag it in every proposal, along with a fallback parts list if the preferred GPU is allocated.