Nokia and NVIDIA are widening their AI-RAN collaboration from reference designs and laboratory work into operator trials and infrastructure build-outs. Nearly 20 operators have publicly announced AI-RAN collaborations built on Nokia's anyRAN software and NVIDIA's Aerial RAN Computer. Nine operators across Europe, the Middle East and Africa are now building or expanding NVIDIA-powered infrastructure, while Nokia is running anyRAN trials with BT, Deutsche Telekom, Elisa and Orange.
A stack built around accelerated radio compute
The reference architecture combines Nokia's anyRAN software with NVIDIA's ARC platform. CUDA, Nemotron open models and Agent Toolkit 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 foundation.
Stacking those elements into one reference design shortens the integration path for operators. It also creates a deeper dependency: the radio layer is tied to NVIDIA's programming model, with Nokia's RAN software running above it. Cisco and Nokia are also folding Nemotron open models into autonomous-network software, extending that dependency into operations tooling.
Why operators are testing distributed inference
The engineering case rests on assets operators already control: land, power, shell space, spectrum and fibre. Adding accelerated compute at centralised, metro and cell sites allows models to run nearer to the machinery, responders and handsets that consume them. It also allows the radio itself to run inference.
One measured example is AI-based channel estimation. Lab measurements from Deutsche Telekom and Nokia put typical uplink throughput improvement at around 10 to 15 percent. That is a useful gain, but a modest return for a distributed compute build-out. Once inference moves outward, backhaul becomes the constraining variable, and Nokia's 800G coherent pluggables indicate the inter-site capacity the company expects operators to need.
The placement question is therefore as important as the compute itself. Centralised sites can aggregate workloads, metro sites can shorten the path to urban customers and cell sites can move inference closest to the radio and the devices it serves. Each step outward changes the balance between latency, local processing and backhaul capacity. Nokia's emphasis on 800G coherent pluggables points to transport as the network element that must scale alongside the compute.
Trials cover a mix of workloads
The early trials are not built around 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.
That workload mix matters. It suggests operators are testing whether one accelerated platform can support consumer traffic, enterprise services and new AI applications. It does not yet show that those applications generate enough revenue to justify widespread deployment.
What the trials do not yet establish is commercial scale. The lab result covers one technique, AI-based channel estimation, and a typical uplink improvement of 10 to 15 percent. Field trials combine voice, data and vision AI, but the disclosed material does not quantify revenue, cost per site or sustained performance across a wide estate. Those gaps matter because the build-out is distributed, not confined to a few data centres.
Vendor positioning and the path to 6G
NVIDIA founder and CEO Jensen Huang said telecom networks were built to carry conversations and will now also carry intelligence. He described operators as turning land, power and connectivity into distributed AI infrastructure, bringing intelligence closer to industries, people and robots, and said the Nokia collaboration extends accelerated computing into the radio network itself as a foundation for AI-native 5G.
Nokia chief technology and AI officer Pallavi Mahajan said the next era of connectivity would be defined by networks that distribute intelligence, continuously optimise performance and enable new AI-powered services, including physical AI. She framed the work with NVIDIA and operators as bringing AI into the radio network and across operations, creating a software-defined foundation intended to help operators get more from spectrum today while opening a path toward autonomous, AI-native 6G.
Economics remain unresolved
The commercial model is still indirect. 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 central to the near-term business case: operators provide the estate, spectrum and financing, while vendors supply the compute platform and software stack.
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 milestones should show whether the reference design can move from lab measurements and limited field trials into repeatable, revenue-generating services.
That makes the 2027 checkpoints more than a schedule. They are a public test of whether a common reference design can support several operator business models at once: BT's live trial, Deutsche Telekom's field work and Elisa's distributed deployment with commercial AI-RAN services. A successful trial would still leave questions about backhaul cost and revenue sharing, but it would show that the integration path is repeatable.
For now, the evidence is a set of operator trials, one completed European field trial, a 10 to 15 percent lab improvement in uplink throughput for AI-based channel estimation and a vendor stack that is still being integrated. The important test is not whether AI can run in the radio network, but whether the resulting services and backhaul economics justify the distributed compute build-out.