These new features expand the possibilities of AI-RAN, an architecture that combines the functions of a radio access network with artificial intelligence capabilities on a software-defined infrastructure. After announcing its first commercial AI-RAN platform in July, Nokia and NVIDIA are now making progress in integrating it into operational networks and preparing for future deployments.

The goal is to provide operators with tools to improve the performance of 5G and 5G Advanced networks, while developing the technological foundations for future AI-native 6G networks.

Artificial intelligence applied to 5G signal processing
One of the main advances of the collaboration consists of introducing artificial intelligence directly into the signal processing chain of 5G networks.

Companies are applying these capabilities to functions such as channel estimation, beamforming, multi-user MIMO pairing, and radio resource planning.

Nokia AI-native anyRAN software, running on NVIDIA Aerial RAN Computer, provides the software-defined foundation for integrating these functions.
The architecture aims to improve network capacity and performance by using artificial intelligence algorithms in radio operations.

Elisa, BT, Deutsche Telekom, and Orange advance in AI-RAN trials.
Collaboration with telecom operators is a central part of Nokia and NVIDIA's work. The announced projects demonstrate different stages of technology evaluation and integration.

Elisa has simultaneously run voice, mobile data, and video analytics services on a single AI-RAN platform across its 5G network in Helsinki. This demonstration has allowed for the combined operation of radio and artificial intelligence workloads in a real-world network environment. The operator plans to move towards a distributed deployment and evaluate customer use cases and commercial AI-RAN services during 2027. BT is preparing a trial on an operational network in Birmingham, also planned for 2027.

For their part, Deutsche Telekom and Nokia are developing AI-based channel estimation functions to improve network performance and efficiency.

Initial laboratory measurements have shown typical improvements of 10 to 15% in uplink performance, with even greater increases under more complex radio conditions. These results are from initial laboratory tests. The next phase will include new spectral efficiency algorithms and their field validation. Orange also plans to advance AI-RAN testing, both in the laboratory and under real-world network conditions.

Spectral efficiency, sensing, and localized inference:
Nokia structures the development of AI-RAN around three main capabilities.
The first is AI-driven spectral efficiency, aimed at obtaining greater capacity from the available radio spectrum.

The second is the integration of sensing and communications, which will allow the use of radio infrastructure to detect and interpret information about the physical environment.

The third is localized inference, which seeks to execute artificial intelligence functions near the place where the data is generated.

These capabilities are part of the evolution toward networks capable of combining connectivity, processing, and environmental analysis. They also form a basis for future applications of physical AI, which require relating digital information to real-world conditions and activities.

Nokia is expanding its AI-RAN ecosystem with distributed applications, called dApps, that can run at the network edge. Nokia's E3 interface provides these applications with real-time access to data on the radio access network
.

This capability allows the RAN infrastructure to be used as a programmable platform for sensing and localized intelligence functions. Among the companies working in this environment is DeepSig, which develops radio spectrum detection software. Its technology uses existing radio infrastructure to identify radio frequency threats, interference, and unauthorized transmitters.

Furthermore, ZaiNar is integrating its positioning software with Nokia Cloud AI-RAN. The goal is to offer precise network-based location capabilities, utilizing existing wireless infrastructure without requiring modifications to connected devices or radio hardware.

Sensing and Communications Demonstration at NVIDIA GTC Berlin:
During NVIDIA GTC Berlin, Nokia presented a demonstration of sensing and communications integration. The demonstration showed how a radio access network can be used to detect and interpret activity in the physical environment.

To facilitate the deployment of these capabilities, Nokia and NVIDIA collaborate with different infrastructure providers.

Dell Technologies, Quanta Cloud Technology, and Supermicro support flexible NVIDIA Aerial RAN Computer configurations for both distributed and centralized deployments. These configurations enable the execution of Nokia 5G anyRAN software alongside AI inference workloads.

Red Hat is also incorporating AI-RAN technologies into a common platform based on Red Hat OpenShift. The goal is to provide a cloud-native infrastructure that enables the execution of radio functions and AI workloads on shared resources.

Nokia and NVIDIA develop an agentic AI platform for autonomous networks.
The collaboration also extends to the automation of telecom operations using agentic artificial intelligence.

Nokia and NVIDIA are integrating NVIDIA AI Enterprise, along with NVIDIA Nemotron models and libraries, into Nokia's agentic platform. This architecture will enable operators to create, evaluate, deploy, and manage AI agents specialized in telecommunications. The agents will be targeted at different network domains, including radio access, core network, transport, and cloud.

Nokia provides the infrastructure knowledge, operational context, and governance mechanisms necessary for agents to interpret network conditions and execute controlled actions. The platform incorporates an autonomy approach called a glass-box, which maintains operator oversight.

It also includes protection mechanisms designed to control how the actions proposed by the agents are tested and executed.

On-premises, cloud, and hybrid deployments with data control.
The joint architecture of Nokia and NVIDIA is designed to support on-premises, cloud, and hybrid deployments. This flexibility allows operators to decide where operational data, AI models, and operating policies are hosted.

The approach aims to develop agentic AI capabilities with control and governance over the resources used. The platform intends to extend artificial intelligence from radio functions to the coordination of operations across different network domains.

Among the operators working with Nokia to identify advanced use cases and establish potential integration timelines are etisalat UAE, MEO, and Airtel. This work is currently in the application definition and integration planning phase.

IP and Optical Connectivity for Distributed AI Infrastructures
In addition to AI-RAN and automation capabilities, Nokia is showcasing IP and optical connectivity solutions at GTC designed to link distributed AI infrastructures. The offering includes technologies for connecting data centers, IP gateways, and wide area networks.

Among the solutions announced are 800G pluggable coherent optical modules, designed to expand connectivity between different data center locations. These technologies are intended to connect radio resources, edge computing, and data centers within a coordinated infrastructure.

With these advances, Nokia and NVIDIA are expanding their collaboration from incorporating artificial intelligence into radio processing to developing distributed applications and progressively autonomous network operations.

Projects with operators, spectral efficiency tests, and the integration of AI agents are part of a technological evolution that seeks to improve current 5G networks and prepare the transition to AI-native 6G architectures.