(Source: SOPA Images Limited/Alamy Stock Photo)

Nokia and SoftBank have agreed to conduct a commercial AI-RAN (AI-Radio Access Network) field trial in Japan in 2027. This marks a notable step forward since the trial will take place in a genuine commercial network environment under real commercial traffic conditions, rather than in a controlled lab or isolated testbed setting.

The trial aims to assess the feasibility of running AI and radio network workloads on shared computing infrastructure simultaneously. The two companies essentially want to determine whether the same physical infrastructure used to handle network traffic could also be used to process AI computing tasks, which could improve overall resource utilization and efficiency.

The trial will see Nokia’s AI-native anyRAN Cloud RAN software combined with SoftBank’s AITRAS core AI-RAN platform. The trial will also heavily rely on Nvidia’s Aerial RAN Computer (ARC) platform. The underlying hardware architecture features Nvidia Grace CPUs, Nvidia RTX PRO 4500 GPUs and Nvidia ConnectX-8 networking cards, which can handle heavy AI and radio workloads simultaneously on a single, shared computing infrastructure.

The management of radio and AI resources will be coordinated by integrating the Nokia MantaRay management platform with the AITRAS Orchestrator.

According to SoftBank, not only are telecom networks increasingly required to accommodate growing data traffic, they are also required to provide the computing capabilities needed for low-latency, real-time AI processing as generative AI, AI agents, and physical AI continue to evolve.

The Japanese operator is developing the AITRAS Orchestrator, a platform designed to run AI and RAN workloads on shared infrastructure. In real time, it will monitor demand for computing resources used for AI processing and RAN control, allocating resources flexibly and dynamically based on factors such as resource availability, application requirements and projected power consumption.

SoftBank said that the field trial will assess the suitability of AITRAS as a common platform for hosting RAN software and AI applications from multiple vendors, without the limitations imposed by a specific RAN software implementation.

"Our field trial with Nokia will mark a significant milestone in moving AI-RAN from concept toward commercialization," Ryuji Wakikawa, SoftBank Corp. senior vice president and CTO, said in a statement. "We will further advance AI-RAN validation toward commercialization and work toward a new era of networks that deliver both connectivity and AI services."

Transition toward AI-native network architecture

In a separate statement, Nokia said that its latest agreement with SoftBank builds on their ongoing collaboration in AI-RAN, which is supported by Nvidia’s accelerated computing node.

The Finnish equipment vendor added that the trial will contribute to the wider industry shift toward AI-native network architectures by exploring how operators can introduce new capabilities more quickly and support the development of new AI applications at the network edge.

The trial will explore how AI-RAN could support richer, AI-driven experiences, lower-latency services, and new applications powered by real-time intelligence. Future use cases could include physical AI, immersive digital experiences, robotics and autonomous systems that require close interaction between AI and connectivity.

Pallavi Mahajan, Nokia's chief technology and AI officer, noted that their ongoing collaboration with SoftBank is a stepping stone toward developing the automated, AI-native network architectures required for future 6G networks.

"By bringing AI and RAN workloads onto common, programmable infrastructure, AI-RAN has the potential to unlock new levels of spectral efficiency while providing the distributed intelligence needed to support physical AI," she said. "As the boundaries between network and compute continue to blur, this distributed infrastructure can become a powerful platform for both connectivity and AI."