Nokia adds NVIDIA-powered AI radio capabilities as operators advance network trials

AI is moving into the equipment that runs mobile networks, not just the services those networks carry. On October 21, 2026, Nokia introduced new radio-network capabilities developed with NVIDIA, alongside operator test results and deployment plans. Elisa has already run mobile traffic and video analytics together on the platform in its live 5G network, Nokia said.

More capacity from the same spectrum

The technology is called AI-RAN: AI applied to the radio access network. Nokia’s anyRAN software runs on NVIDIA’s Aerial RAN Computer, providing a software-defined platform for radio functions and AI workloads. The new capabilities target both the work of handling mobile signals and running AI close to where data is generated.

Nokia and NVIDIA are applying AI within several parts of 5G radio processing, including channel estimation, beamforming, pairing users for multi-user antenna transmissions, and scheduling. Nokia’s stated goal is to extract more capacity from existing spectrum. That makes radio performance a distinct part of the announcement, separate from hosting additional AI applications on network infrastructure.

With Deutsche Telekom, Nokia is developing AI-based channel estimation as one component of that effort. The initial performance measurements came from laboratory testing, not a completed field evaluation. The next phase will add more AI algorithms aimed at improving spectrum efficiency, followed by field validation.

10–15%Typical uplink throughput improvement in initial lab measurements

Nokia reported this gain from its AI-based channel-estimation work with Deutsche Telekom, with larger improvements under more complex radio conditions. The result concerns uplink traffic and one radio-processing technique, not a measured improvement across an entire deployed network.

Operators are at different stages

Elisa’s Helsinki work demonstrated simultaneous voice, mobile data and video analytics on the same AI-RAN platform. The next steps remain prospective: distributed deployment and evaluation of customer use cases and commercial AI-RAN services are expected in 2027. The live-network demonstration and those future commercial evaluations are different milestones.

BT is preparing a live-network trial in Birmingham in 2027. Orange plans testing in both laboratory and real-network conditions. Together with Deutsche Telekom’s work, these programs cover different stages of development: laboratory measurements, live-network testing and preparation for deployment. Nokia presents them as progress toward commercial use, rather than a completed rollout across these operators.

A related milestone came at T-Mobile’s commercial cell site in Bellevue, Washington. In a same-day account, Nokia said a call ran over T-Mobile’s production network using Nokia radio equipment and anyRAN software, NVIDIA Aerial RAN Computer infrastructure, a commercial 5G core and commercial user equipment.

For that program, trials of additional AI radio algorithms and AI inference—running trained models—alongside radio functions are still ahead. The distinction is important: a successful radio call on accelerated computing infrastructure is not the same demonstration as running radio and AI application workloads together.

Applications reach into the radio network

Nokia is also extending the platform through distributed applications, or dApps, that run at the network edge. Its E3 interface gives those applications real-time access to radio-network data. This supports another part of the announcement: using wireless infrastructure for sensing and local intelligence, rather than only transporting traffic.

• DeepSig is developing spectrum-sensing software that uses existing radio infrastructure to identify radio-frequency threats, interference and rogue transmitters.
• ZaiNar is integrating positioning software with Nokia Cloud AI-RAN for network-based location capabilities. Nokia says the approach requires no changes to connected devices or radio hardware.

The deployment ecosystem includes Dell Technologies, Quanta Cloud Technology and Supermicro, which support distributed and centralized Aerial RAN Computer configurations running Nokia’s 5G software alongside AI inference. Red Hat is bringing AI-RAN technologies onto OpenShift, its cloud-native platform. Those integrations address the shared infrastructure needed to host radio functions and AI workloads as operator testing advances.