AI-RAN needs to demonstrate that it can support AI services alongside mobile connectivity under real-world network conditions.
In an Oct. 20 announcement, Nokia said it and SoftBank plan to conduct an AI-RAN field trial in Japan in 2027. The project will combine Nokia’s anyRAN software with SoftBank’s AITRAS platform in a commercial network environment under commercial traffic conditions.
AI-RAN combines artificial intelligence with radio access network infrastructure. For telecom operators, the opportunity is to use infrastructure for both connectivity and AI processing. The challenge is maintaining dependable mobile services while sharing computing resources.
“Our planned field trial with Nokia will mark a significant milestone in AI-RAN’s transition from concept to execution. We will integrate Nokia’s AI-native anyRAN into our AITRAS platform as its RAN component, with the AITRAS Orchestrator working in coordination with Nokia’s RAN and MantaRay,” said Ryuji Wakikawa, senior vice president and CTO at SoftBank Corp.
SoftBank’s efforts toward AI-RAN
The Nokia agreement builds on SoftBank’s earlier AI-RAN collaborations.
In February, Ericsson and SoftBank announced interworking between the AITRAS Orchestrator and Ericsson Intelligent Automation Platform, which manages and controls radio access networks. The demonstrated setup coordinated RAN automation applications and AI workloads across separate compute clusters, enabling dynamic resource allocation without placing both on the same server.
In March, Mitsubishi Heavy Industries (MHI) and SoftBank announced a collaboration involving an AITRAS field trial at MHI’s Yokohama Hardtech Hub. The planned test would analyze smartphone images of defective products to identify damaged areas and determine repair methods. It would operate in an on-premises edge data center isolated from external networks.
The planned Nokia trial extends that work to commercial network traffic, where coordinating AI workloads must also preserve mobile network operating requirements.
What eWeek found: AI-RAN’s business case still needs testing
The partner ecosystem addresses an important implementation challenge: AI-RAN requires computing hardware, network software and workload management to function together. For enterprise readers, however, the more useful question is what that integration could deliver in reliability, response times and cost.
SoftBank’s AI-RAN collaborations span computing hardware, cloud infrastructure, network software and industrial applications, reflecting the broader need to integrate compute, networking and software for enterprise AI.
The planned Nokia trial will use an Nvidia Grace CPU, an Nvidia RTX PRO 4500 GPU and Nvidia ConnectX-8 networking.
For cloud infrastructure, SoftBank has worked with Red Hat. In March 2025, the companies described AITRAS as an integrated AI and RAN solution built on Red Hat OpenShift, a Kubernetes-based application platform. This provides a common cloud-native software environment for virtualized RAN functions and AI applications. Their collaboration also includes power monitoring to support workload placement and energy optimization.
The partners serve different roles: Nokia contributes RAN software and management integration, Ericsson contributes network automation integration, and MHI provides an industrial application use case.
For enterprises considering AI services delivered through telecom infrastructure, the potential benefit is processing closer to users and connected equipment. MHI’s planned inspection application illustrates one possible use: analyzing operational images locally while keeping confidential product data within a closed environment. That example should not be taken as proof that the same capabilities are already available across a public mobile network.
Shared infrastructure also creates a resource-management question. Operators need to accommodate changing AI demand while preserving mobile network performance. Useful trial results would show how the platform handles competing workloads, traffic peaks and service interruptions.
The economics will matter alongside performance. Operators will need to weigh equipment, energy and management costs against demand for AI processing. Enterprise customers will need clear pricing, data-handling terms and service commitments before comparing an operator’s offering with cloud or on-premises alternatives.
The 2027 trial is therefore a milestone to watch, rather than a confirmed production rollout. Evidence of sustained reliability and economic value will determine how compelling AI-RAN becomes for operators and their customers.
Read more: Equinix’s planned Inference Exchange highlights the pricing, deployment and service details enterprises need when evaluating AI processing closer to users and data.