T-Mobile has completed the world’s first AI-RAN call on a live commercial network, NVIDIA and Nokia announced at GTC Berlin on Wednesday — a working phone call carried by a radio access network running on GPU-accelerated computing, in Bellevue, Washington, rather than in a lab.
The call is the headline of a wider telecom push. Nine operators across Europe, the Middle East and Africa are building or expanding NVIDIA-powered infrastructure, Nokia is taking its anyRAN software on NVIDIA’s Aerial RAN Computer into trials at BT, Deutsche Telekom, Elisa and Orange, and Nokia and Cisco are folding NVIDIA’s Nemotron open models into the software that runs autonomous networks.
The First AI-RAN Call: Bellevue, On A Production Network
T-Mobile’s call ran on commercial Aerial RAN Computer servers, Nokia’s NVIDIA CUDA-accelerated RAN software and massive MIMO radios, connected to an existing 5G Standalone core and an ordinary commercial smartphone. That detail matters: AI-RAN demonstrations have usually lived in labs or on test handsets, and this one used the carrier’s production setup.
Europe’s first field trial belongs to Elisa, which ran voice, mobile data and video analytics simultaneously on one accelerated platform over 5G spectrum in an over-the-air field environment — NVIDIA says the trial clears the way for commercial AI-RAN services in 2027. BT is planning live trials in Birmingham, England, in 2027. Deutsche Telekom validated the Nokia-NVIDIA platform in its Bonn and Oulu labs, including AI-enhanced uplink channel estimation that improved uplink throughput, with field trials planned for 2027. Orange is testing Nokia’s 5G software on the platform for spectral efficiency and edge services.
NVIDIA also introduced two new Aerial RAN Computer configurations, MGX 2U and MGX 4U, for dense, high-capacity radio and intelligence workloads at centralised RAN sites. They run on RTX PRO 6000 Blackwell Server Edition GPUs with x86 or Vera CPU options, and Dell Technologies, Quanta Cloud Technology and Supermicro will offer ARC-ready servers. Nearly 20 operators worldwide have now publicly announced AI-RAN collaborations on Nokia anyRAN and NVIDIA ARC.
Nine EMEA Operators Build The Distributed Layer
The infrastructure list is long. New NVIDIA-powered builds are under way at CRA in the Czech Republic, e& in the United Arab Emirates, Silknet in Georgia and TIM in Italy. Expansions are running at Cassava in Africa, Deutsche Telekom in Germany, Fastweb + Vodafone in Italy, Kazakhtelecom in Central Asia and Orion Telecom in Serbia.
Deutsche Telekom plans to expand its NVIDIA-powered Industrial AI Cloud, which hosts open models including Nemotron and Soofi for robotics customers such as Agile Robots. Orange is hosting open models, including Nemotron, as a managed service on its Live Intelligence platform, and is deploying ARC hardware for voice intelligence — using Nemotron Speech to help emergency and field teams translate across languages in real time.
On the operations side, Nokia is integrating Nemotron into its Autonomous Networks Suite with the NVIDIA Agent Toolkit, Cisco has fine-tuned Nemotron models inside Crosswork AI, and TCS, NTT DATA and BT are building network agents on the same stack — agents that validate radio-site builds, spot degradation and inspect field-installation photos.
Why It Matters
“Telecom networks were built to carry the world’s conversations. Now, they will also carry the world’s intelligence,” said Jensen Huang, NVIDIA’s founder and CEO. The economics behind the line are the story: operators own the land, power, fibre and spectrum that AI infrastructure is short of, and AI-RAN turns their cell sites from a cost centre into rentable compute near users. For NVIDIA, the radio network is a new GPU market measured in hundreds of thousands of sites, and it extends the same open-model strategy behind releases such as the Nemotron 3.5 open models into software operators already run. For the industry, a commercial-network first call moves AI-RAN from conference concept to procurement question — the pattern GEroundtable follows in why NVIDIA powers the AI boom. It also gives the agent-safety story a new venue: the networks these agents will run on are being rebuilt around the same stack as NVIDIA’s OpenShell architecture for governing them.
Background
AI-RAN — running the radio access network, the part of a mobile network that talks to phones, on general-purpose accelerated computing instead of fixed-function hardware — has been NVIDIA and Nokia’s joint project for several years, part of a broader industry push toward AI-native 5G and eventually 6G. Until now the evidence was lab results and field trials. A completed call on T-Mobile’s live commercial network, with European operators queuing trials behind it, is the first proof point aimed squarely at operators deciding their next RAN contracts.
Source: NVIDIA via GlobeNewswire, Unite.AI
0 Comments