Nokia’s AI-RAN Platform: A Radio Comeback That Runs on NVIDIA

Nokia and NVIDIA have unveiled a collaborative AI-RAN (Radio Access Network) architecture designed to integrate machine learning directly into cellular infrastructure. Built on Nokia’s proprietary anyRAN software framework and NVIDIA’s Aerial GPU platform, the solution allows telecom operators to run both network workloads and AI models on the same server hardware. This launch marks a major shift from custom, proprietary network hardware toward standard GPU-accelerated computing.

### How Nokia Integrates AI into Radio Networks Nokia uses AI algorithms to optimize signal transmission and frequency allocation in real time. Instead of relying on static rules, Nokia’s software uses machine learning models for predictive beamforming—shaping radio signals to target moving users precisely—and dynamic traffic routing. The core of this system is Nokia’s anyRAN framework, which decouples the radio processing software from underlying server hardware. This software-driven design allows telecom operators to deploy network functions as software packages on cloud-native servers, dynamically tuning frequencies and cutting interference. Pilot testing has already shown immediate spectral efficiency gains of over 20%, with mid-term projections targeting a 50% efficiency boost by 2027 and a 100% capacity increase by 2028.

### How NVIDIA Accelerates and Powers Cellular AI NVIDIA supports this architecture through its specialized Aerial SDK, a software-defined 5G platform. Instead of using custom, expensive silicon chips (ASICs) built specifically for cellular processing, Nokia’s platform offloads Layer 1 (physical layer) signal processing to standard NVIDIA enterprise GPUs. By using NVIDIA’s CUDA software ecosystem, the GPUs can dynamically balance processing tasks. During periods of low network traffic, operators can shift the unused GPU compute power to host AI edge workloads—such as smart city analytics, factory computer vision, or real-time security scanning. This multi-tenant capability transforms base stations from single-purpose signal routers into high-performance edge-computing centers.

### Nokia’s Latest Enterprise AI Initiatives Beyond the AI-RAN alliance, Nokia is investing in several key artificial intelligence projects: 1. **Nokia AVA Cognitive Services:** A cloud-based software suite that uses machine learning to analyze network telemetry. AVA predicts component failures before they happen, automates energy saving by putting idle radio cells to sleep, and detects security threats. 2. **Nokia MX Industrial Edge (MXIE):** An on-premise compute platform that hosts operational applications for smart factories. Integrated with private 5G, MXIE allows industrial companies to run real-time AI computer vision models and autonomous mobile robot (AMR) control scripts directly at the factory floor. 3. **The Global AI-RAN Alliance:** Nokia is a founding member of this group alongside NVIDIA, Arm, SoftBank, and T-Mobile, collaborating to standardize AI-RAN architectures across the telecom industry.

### The Strategic Business Impact & Market Rivalry This shift is a key part of CEO Justin Hotard’s plan to restore profitability to Nokia's Mobile Infrastructure business. By partnering with NVIDIA—which recently invested $1 billion for a 3% stake in the network provider—Nokia is outsourcing costly silicon R&D to focus on high-margin software subscriptions. While research firm Omdia estimates the total AI-RAN market size will exceed $200 billion by 2030, Nokia faces direct competition. In June, Ericsson launched an alternative AI-in-RAN product that works on operators' existing baseband processors without needing GPU hardware, claiming up to 20% higher download speeds. The core debate between Nokia's high-performance NVIDIA GPU dependency and Ericsson's hardware-agnostic software approach will shape the future of cellular infrastructure.