Nokia and NVIDIA are moving AI-RAN from reference architecture to operator trials, but the disclosed results and deployment schedule suggest that the commercial case is still being assembled. Nine operators across Europe, the Middle East and Africa are now building or expanding NVIDIA-powered infrastructure, while nearly 20 operators have publicly announced AI-RAN collaborations built on Nokia's anyRAN software and NVIDIA's Aerial RAN Computer.
Nokia is running anyRAN trials with BT, Deutsche Telekom, Elisa and Orange. The reference design combines Nokia's radio software with the Aerial RAN Computer, CUDA, Nemotron open models and an Agent Toolkit. Those AI and automation 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, and Red Hat OpenShift provides the cloud-native layer beneath them.
The distinction between announced collaborations and deployed infrastructure matters. The nearly 20 figure covers public announcements, while the nine-operator figure refers to operators now building or expanding NVIDIA-powered infrastructure, and Nokia's named anyRAN trials cover four operators. The evidence for commercial traction therefore rests on trials and infrastructure commitments rather than disclosed AI-service revenue.
The engineering bet: compute at the radio
The engineering case rests on assets operators already control at the land, power and shell layer, plus spectrum and fibre. Adding accelerated compute at centralised, metro and cell sites allows operators to serve models closer to the machinery, emergency responders and handsets that consume them. It also allows the radio itself to run inference. That is a different proposition from adding AI in a distant cloud: the network becomes part of the inference path, not merely a transport link.
Deutsche Telekom and Nokia report lab measurements showing typical uplink throughput improvement of around 10 to 15 percent for AI-based channel estimation. The figure is specific, but it is also modest for a distributed compute build-out. It addresses one radio function under lab conditions, not a portfolio of revenue-generating AI services. The gap between a 10-15 percent uplink gain and a return on compute at many cell sites is the central question in the trials.
Backhaul and workload mix set the boundaries
Backhaul becomes the constraining variable once inference moves outward. Nokia's 800G coherent pluggables give some indication of the capacity the company expects operators to need between sites. The detail matters because the distributed compute proposition depends on moving both user traffic and model-related data across the same wide-area network.
The trials cover a mix of workloads, not a single radio feature. 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. Each use case tests a different combination of latency, throughput and operational integration, which makes a single benchmark insufficient to judge the architecture.
A reference design with platform dependencies
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, extending that dependency into operations tooling. The benefit is a more defined starting point for trials; the cost is a narrower set of platform choices at the radio and operations layers.
Jensen Huang, founder and CEO of NVIDIA, framed telecom networks as systems built to carry conversations that will now also carry intelligence. He said leading operators are turning land, power and connectivity into distributed AI infrastructure, bringing intelligence closer to industries, people and robots. Pallavi Mahajan, Nokia's chief technology and AI officer, said the next connectivity era would distribute intelligence, continuously optimise performance and enable AI-powered services, including physical AI, with AI-native 6G as a longer-term path.
Who pays before AI revenue arrives
Until AI-service revenue materialises, operators carry the financing for distributed compute and backhaul. Suppliers of accelerated compute and RAN software hold the layer where incremental value would accumulate. That division is visible in the reference design: operators bring sites, spectrum and fibre; Nokia brings RAN software; NVIDIA brings accelerated computing and its programming model; server and cloud suppliers fill out the stack. This is not a verdict on demand, but it explains why the trials emphasise integration and workload diversity as much as raw performance.
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 matter because they separate announced collaborations from operational services. They also set the next test: whether distributed compute at centralised, metro and cell sites can produce enough radio efficiency and new service revenue to justify the build-out.