Nvidia and Nokia Bring AI Into Live Mobile Networks

Nvidia and Nokia are moving their AI ambitions deeper into mobile networks. At GTC Berlin on 21 October, Nvidia said T-Mobile had completed an AI-RAN call on a commercial network in Bellevue, Washington. The call is a technical milestone, while the wider promise of AI-powered radio networks remains to be tested through larger deployments. BT is preparing live trials in Britain in 2027, and other operators are running or planning their own pilots.

AI-RAN stands for artificial intelligence radio access network. The radio access network is the part of a mobile system that connects phones to towers and passes traffic onward. In Nvidia and Nokia’s approach, accelerated computing can run some radio software and AI workloads on a shared platform. The companies argue that this could help operators use spectrum more efficiently and deliver services close to customers without building a completely separate edge-computing network.

According to Nvidia’s announcement, the T-Mobile call used commercial Aerial RAN Computer servers, Nokia’s accelerated RAN software and radios, an existing standalone 5G core and a commercial smartphone. Nvidia calls it the first AI-RAN call on a commercial network. That description speaks to the test environment and equipment; it does not mean AI-RAN has been deployed across T-Mobile’s national network.

Europe is at an earlier and varied stage. Finland’s Elisa says it completed a field trial that ran voice, mobile data and video analytics together on one accelerated platform over 5G. BT is planning trials in Birmingham in 2027. Deutsche Telekom has validated the technology in laboratories and aims to move to field trials next year. Orange is testing Nokia software on Nvidia hardware for network efficiency and edge services. These are different levels of progress, so grouping all of them as commercial launches would overstate the announcement.

The commercial question is whether shared infrastructure can deliver better performance or lower cost than the equipment operators already use. Radio networks are built for stringent reliability and latency requirements. Adding AI applications to the same computing system introduces new demands for power, security and capacity planning. Operators will need to prove that an inference workload cannot interfere with the phone call or data session the network exists to carry.

The announcement also places African infrastructure in a wider strategy. Nvidia says Cassava Technologies is expanding its AI-powered capacity across the continent, one of nine operator groups in Europe, the Middle East and Africa building or expanding Nvidia-backed infrastructure. That is separate from the T-Mobile AI-RAN call. It should not be read as evidence that an African mobile network has already deployed the same radio architecture.

TechBooky has previously reported on Cassava’s African AI factory plans. For telecoms groups that already control towers, fibre, data centres and customer relationships, Nvidia sees an opportunity to turn those assets into a platform for local AI services. In African markets, the appeal could include lower-latency applications and more control over where data is processed. The hard constraints remain electricity, backhaul, capital costs and whether enough customers will pay for the new services.

Nvidia also introduced two configurations of its Aerial RAN Computer for denser network sites. Nokia plans to offer its anyRAN software on the platform, while BT, Deutsche Telekom, Elisa and Orange are among the trial partners. Separately, Nokia and Cisco are incorporating Nvidia Nemotron models into network-management tools, aiming to automate tasks such as fault diagnosis and service operations.

The story is therefore bigger than a single demonstration call, but it is not yet a finished transformation of mobile networks. Over the next year, the meaningful measures will be field-trial results, energy use, equipment costs and the number of sites operators can run reliably. For customers, the technology matters only if it leads to better service or useful applications rather than another expensive layer inside the network.