Nokia Puts AI Into the 5G Signal Chain With NVIDIA, and Six Operators Are Already Testing It

Nokia and NVIDIA have advanced their AI-RAN collaboration toward commercial deployment, integrating AI into the 5G radio signal processing chain for channel estimation, beamforming, multi-user MIMO pairing and scheduling. The announcement came at NVIDIA’s GTC conference in Berlin on October 26, 2026, roughly a year after NVIDIA invested one billion dollars in the Finnish vendor. Nokia’s AI-native anyRAN software runs on the NVIDIA Aerial RAN Computer, creating a software-defined framework for AI-driven radio functions.

Six operators are already testing or deploying the technology. T-Mobile has completed a live-network AI-RAN call in the United States. Elisa is running simultaneous voice, data and video workloads on AI-RAN servers on its live network in Finland. Deutsche Telekom is piloting AI-RAN at distributed sites on its live network in Germany and has recorded uplink gains of 10 to 15 percent using AI-based channel estimation in lab tests, expanding to field trials in 2027. Orange is testing AI-RAN for spectral efficiency and edge services in France. BT will do the same in the UK next year.

NVIDIA says its software-defined RAN platform, using Nokia’s RAN software, now outperforms traditional proprietary RAN systems on performance per watt. The claim is that AI-led spectral efficiency maximises capacity from existing spectrum resources, while integrated sensing and communication helps networks comprehend their physical surroundings. Localised inference processes intelligence nearer to data sources.

The three key capabilities form part of a larger platform: AI-led spectral efficiency, integrated sensing and communication, and localised inference. Nokia is also broadening its AI-RAN ecosystem through decentralised applications that operate at the network edge, using the E3 interface to give applications real-time access to RAN data. This turns the radio network into a programmable platform for sensing and localised intelligence.

Nine EMEA operators are building or expanding NVIDIA infrastructure beyond the RAN trials. Deutsche Telekom is extending its Industrial AI Cloud for enterprise AI and robotics. Jensen Huang, NVIDIA’s chief executive, predicted that every telco with 100,000 sites representing some 100 GW of distributed base stations and data centres will evolve into a distributed AI network. The 100 GW figure is not easily verified and appears to be an estimate of future AI capacity, not an audited survey.

The collaboration competes directly with Ericsson’s approach. Ericsson’s chief executive Per Narvinger has defended custom ASIC-based AI RAN, citing 10 percent spectrum-efficiency gains from AI-native link adaptation across more than 60,000 base stations with three operators. Ericsson keeps GPU options open but argues that custom silicon delivers functionality faster today. The three major vendors are now competing on architecture philosophy as much as on performance metrics.

GSMA data on 5G adoption provides the market context. The Mobile Economy reports track 5G subscriptions, which reached 3.3 billion globally, and the number of live 5G networks with network slicing, which grew from 65 to 84 in the past year. These figures frame the AI-RAN push as part of a broader shift toward software-defined networks that can support differentiated services beyond basic connectivity.

Nokia has not disclosed the number of sites running AI-RAN in production, the cost of the NVIDIA Aerial RAN Computer hardware, or whether the performance-per-watt claim has been independently benchmarked. The company also did not release revenue figures attributable to the AI-RAN product line.

Supplier’s note: We hold Nokia base station equipment and optical transport units from the 2025 production batch in Shenzhen stock, roughly 40 units across five SKUs, with lead time of six to eight working days on fresh orders. Cross-reference stock includes compatible power amplifiers, filters and mounting brackets from third-party vendors that fit the same tower configuration. Customers typically order in lots of two to four units for site upgrades, and we see repeat orders from the same system integrators every five to seven weeks. Pricing holds for 30 days from quotation. We also stock spare fans, power supplies and cable assemblies for the same chassis families, and can arrange on-site installation support through our partner network in Southeast Asia and the Middle East region.