Nokia and NVIDIA are moving AI-RAN from reference designs into operator trials, but the commercial case remains a work in progress. Nine operators across Europe, the Middle East and Africa are now building or expanding NVIDIA-powered infrastructure, according to the companies, and Nokia is running anyRAN trials with BT, Deutsche Telekom, Elisa and Orange. Nearly 20 operators have publicly announced AI-RAN collaborations built on Nokia's anyRAN software and NVIDIA's Aerial RAN Computer.
The phrase AI-RAN covers a broad set of workloads. In this deployment model, accelerated compute sits at centralised, metro and cell sites, allowing operators to run AI inference closer to the machinery, emergency responders and handsets that consume it. It also allows the radio itself to run inference. That is a different proposition from adding a single AI feature to a radio; it is an attempt to turn distributed telecom assets into general-purpose AI infrastructure.
The stack behind the trials
The reference architecture combines Nokia's anyRAN software with NVIDIA's Aerial RAN Computer (ARC). CUDA, Nemotron open models and Agent Toolkit are embedded into Nokia's Autonomous Networks Suite and into third-party operations software from Cisco, TCS and NTT DATA. Dell, QCT and Supermicro supply ARC-ready servers, while Red Hat OpenShift provides the cloud-native foundation beneath them.
For operators, the appeal is partly architectural. They already control land, power, shell space, spectrum and fibre. Adding accelerated compute at existing sites could let them serve AI models nearer to industrial equipment, public-safety users and mobile devices without building entirely new facilities. Nokia and Deutsche Telekom report lab measurements showing typical uplink throughput improvement of about 10% to 15% for AI-based channel estimation.
That is a useful gain, but a modest one relative to the scale of a distributed compute build-out. The figure applies to a specific radio function, not to the entire AI-RAN proposition. As inference moves outward, backhaul becomes the constraining variable. Nokia's 800G coherent pluggables signal the capacity the company expects operators to need between sites, but they also underline that compute at the edge is only as useful as the transport behind it.
A workload mix, not one feature
The trial portfolio shows how broadly operators are interpreting the opportunity. Elisa completed Europe's first AI-RAN field trial, running voice, mobile data and vision AI together on one accelerated platform over 5G. Orange is introducing voice intelligence services for enterprises and emergency responders. Deutsche Telekom is working on physical AI for industrial customers, and BT is preparing live trials in the UK. This is a workload mix, not a single radio feature.
Those use cases matter because they begin to describe where AI-RAN revenue could come from. Vision AI, voice intelligence and industrial physical AI are services that could be sold to enterprises or public agencies. Yet the disclosed material does not include pricing, contract values or customer commitments. The trials establish technical feasibility and operator interest, not commercial demand.
The vendor-layer trade-off
Stacking anyRAN, ARC, CUDA and Nemotron into one reference design shortens the integration path. It also ties operators to NVIDIA's programming model at the radio layer, with Nokia's RAN software sitting above it. Cisco and Nokia are both folding Nemotron open models into the software used to run autonomous networks, which extends that dependency into operations tooling.
That is a strategic trade-off. A tightly integrated stack can reduce time-to-market and give operators a single engineering target. It can also narrow their options at the layer where AI models, accelerators and radio software meet. Nokia's position would be stronger if operators adopt its anyRAN layer as the radio-facing software, while NVIDIA gains a larger footprint in the network's compute base. The suppliers of accelerated compute and RAN software would hold the layer where incremental value accumulates.
Financing and the 2027 checkpoints
The financing question is unresolved. Until AI-service revenue materialises, operators carry the cost of distributed compute and backhaul. That is a significant commitment for infrastructure that must be deployed before demand is proven. The disclosed checkpoints cluster in 2027: BT's Birmingham live trial is slated for then, alongside Deutsche Telekom's field trials and Elisa's distributed deployment with commercial AI-RAN services.
Nokia's Chief Technology and AI Officer Pallavi Mahajan said the next era of connectivity will be defined by networks that distribute intelligence, continuously optimise performance and enable AI-powered services, including physical AI. She described the work with NVIDIA and operators as a software-defined foundation for getting more from spectrum and a practical path toward autonomous, AI-native 6G.
NVIDIA founder and CEO Jensen Huang said telecom networks were built to carry the world's conversations and will now also carry its intelligence. He said leading operators are turning land, power and connectivity into distributed AI infrastructure, with Nokia extending accelerated computing into the radio network itself.
Those claims set out the ambition, but the evidence so far is a set of trials, a reference design and one modest lab measurement. AI-RAN may become a new layer of operator infrastructure, or it may remain a specialised complement to conventional RAN. The 2027 checkpoints will be the first real test of whether distributed AI compute can be financed, operated and sold at scale.