Nokia and NVIDIA are moving AI-RAN from reference designs into operator trials, with Nokia running anyRAN trials with BT, Deutsche Telekom, Elisa and Orange. Nine operators across Europe, the Middle East and Africa are now building or expanding NVIDIA-powered infrastructure, and nearly 20 operators have publicly announced AI-RAN collaborations built on Nokia’s anyRAN software and NVIDIA’s Aerial RAN Computer.
The architecture stacks Nokia’s anyRAN software on the Aerial RAN Computer with CUDA, Nemotron open models and Agent Toolkit. Those components are embedded in Nokia’s Autonomous Networks Suite and in 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. The combination is intended to shorten integration work by packaging radio software, accelerated compute and AI models into one reference design.
Where the compute goes
The engineering case rests on assets operators already control: land, power and shell space at network sites, plus spectrum and fibre. Adding accelerated compute at centralised, metro and cell sites lets operators serve AI models nearer to the machinery, responders and handsets that consume them. It also allows the radio itself to run inference. In that model, AI-RAN is not a single radio feature but a workload mix that can include voice, mobile data, vision AI and other services on one accelerated platform.
The placement options matter. Centralised, metro and cell sites bring compute nearer to different parts of the network, from aggregation points to the radio edge. The further inference moves outward, the more backhaul capacity and site engineering become part of the cost.
Nokia and Deutsche Telekom lab measurements put typical uplink throughput improvement for AI-based channel estimation at around 10 to 15 percent. That is a modest return for a distributed compute build-out. Once inference moves outward, backhaul becomes the constraining variable. Nokia’s 800G coherent pluggables signal the capacity the company expects operators to need between sites, even if the initial AI-RAN gains are incremental.
Trials test different workloads
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. These are different use cases, which matters because the commercial case for AI-RAN will depend on whether operators can fill the same compute footprint with several revenue-generating workloads rather than one radio optimisation.
NVIDIA founder and CEO Jensen Huang said telecom networks were built to carry the world’s conversations and will now also carry the world’s intelligence. He said leading operators are turning land, power and connectivity into distributed AI infrastructure, bringing intelligence closer to industries, people and robots, and that working with Nokia extends accelerated computing into the radio network itself to create a foundation for AI-native 5G.
Pallavi Mahajan, Nokia’s chief technology and AI officer, said the next era of connectivity will be defined by networks that do more than connect: they will distribute intelligence, continuously optimise performance and enable new AI-powered services, including physical AI. She said Nokia, NVIDIA and operators are bringing AI into the radio network and across network operations, creating a software-defined foundation that helps operators get more from spectrum today while opening a path toward autonomous, AI-native 6G.
Benefits come with platform dependence
Stacking anyRAN, ARC, CUDA and Nemotron into one reference design shortens the integration path, but 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, extending that dependency into operations tooling. Operators therefore gain a more integrated route to AI-RAN, while accepting a narrower set of platform choices at a sensitive layer of the network.
The model also changes supplier roles: Nokia provides RAN software, NVIDIA supplies accelerated compute and the programming model, server vendors supply ARC-ready hardware, and Red Hat supplies the cloud foundation. Value may accrue to whoever controls the accelerated-compute layer and the software that manages it.
The financing question is equally important. Until AI-service revenue materialises, operators carry the cost of distributed compute and backhaul, while suppliers of accelerated compute and RAN software hold the layer where incremental value would accumulate. 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.
Those dates make the next phase a test of execution rather than vision. The early signals are real: named operators, lab measurements, a multi-vendor stack and trials that combine several workloads. The unresolved issues are whether backhaul economics improve, whether AI services generate enough revenue to justify the build-out, and whether operators can avoid excessive dependence on a single accelerated-compute platform. Nokia and NVIDIA have moved AI-RAN closer to live networks; the business case still has to be proved in them.