Skip to content
Neural Daily

BusinessNews

AI-RAN: telcos put NVIDIA GPUs into mobile networks

At GTC Berlin, NVIDIA and Nokia said T-Mobile completed the first AI-RAN call on a commercial network. Deutsche Telekom, BT, Elisa and Orange are starting AI-RAN pilots.

By: Mikola Máté Patrik · · 2 min read

NVIDIA press image: NVIDIA-Powered Telecom Networks
Image: NVIDIA · Press kit · Official NVIDIA press/blog image · Source

What changed?

On October 21 at GTC Berlin, NVIDIA announced that major telecom operators are adding accelerated computing to central data centers, metro sites and cell sites (NVIDIA Newsroom), so models and applications can run closer to users.

AI-RAN goes live

  • T-Mobile completed the first AI-RAN call on a commercial network in Bellevue, Washington, using NVIDIA Aerial RAN Computer (ARC) servers and Nokia's CUDA-accelerated RAN software.
  • Elisa ran Europe's first AI-RAN field trial, with voice, mobile data and video analytics running together on one platform over 5G spectrum. Commercial service is planned for 2027.
  • Deutsche Telekom validated the Nokia and NVIDIA platform in labs, with field trials in 2027. BT plans live trials in Birmingham in 2027, and Orange is testing spectral efficiency.
  • New hardware: two ARC configurations (MGX 2U and MGX 4U) with RTX PRO 6000 Blackwell GPUs and x86 or Vera CPU options. Dell, QCT and Supermicro will offer servers.

Nemotron for network operations

Nokia is adding Nemotron to its Autonomous Networks Suite and Cisco to Crosswork AI, both alongside Agent Toolkit. TCS, NTT DATA and BT are building agents to detect network issues and check field-installation photos.

New capacity in the region

Nine EMEA operators are building or expanding NVIDIA infrastructure. New: CRA (Czech Republic), etisalat (UAE), Silknet (Georgia) and TIM (Italy). Expanding: Cassava, Deutsche Telekom, Fastweb + Vodafone, Kazakhtelecom and Orion Telecom (Serbia).

Why it matters

Telcos already own sites, power and spectrum, and now want to add AI capacity on top. If it works, low-latency inference could come from the network edge rather than only from hyperscalers. Commercial scale will be tested in 2027 at the earliest.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles