AI-RAN: NVIDIA Powers Next-Gen Wireless Networks at MWC 2026

Software Takes Center Stage in Next-Generation Wireless

The move toward software-defined, AI-enhanced radio access networks (AI-RAN) is gaining significant momentum, with NVIDIA and its partners demonstrating tangible progress in bringing this technology from laboratory testing to real-world deployments. This shift represents a fundamental change in how wireless networks are built and operated, moving away from dedicated hardware to a more flexible, adaptable, and intelligent software-based approach. Recent collaborations and benchmark results, unveiled ahead of Mobile World Congress (MWC) in Barcelona, showcase the viability of this architecture for both 5G and the emerging 6G standards.

From Concept to Commercial Deployment

Several major telecommunications operators are now actively implementing NVIDIA-powered AI-RAN platforms. T-Mobile U.S., for example, has successfully demonstrated concurrent AI and RAN processing using Nokia’s CUDA-accelerated RAN software. In field tests, Nokia’s AirScale massive MIMO radio operating in the 3.7GHz band supported commercial devices running applications like video streaming, generative AI, and AI-powered video captioning alongside standard 5G services. NVIDIA is showcasing these advancements at MWC, alongside numerous partners.

SoftBank achieved a milestone with its AITRAS live field trial, realizing a 16-layer massive MIMO system using fully software-defined 5G on NVIDIA’s AI-RAN platform. This represents a significant technical step toward commercializing AI-RAN. Indosat Ooredoo Hutchison (IOH) has also progressed from proof-of-concept to pre-commercial field validation, implementing software-defined 5G with Nokia’s vRAN software on NVIDIA AI-RAN platforms. A demonstration at MWC featured Southeast Asia’s first AI-powered 5G call, enabling secure, real-time cross-border connectivity and remote control of a robotic dog over a live 5G network.

Benchmarking Performance and Efficiency

New benchmarking data from SynaXG highlights the performance capabilities of AI-RAN running on NVIDIA platforms. The tests demonstrated high-speed, carrier-grade performance – meaning a high degree of reliability – across multiple 5G spectrum bands. Notably, SynaXG achieved a throughput of 36 Gbps with under 10 milliseconds of latency, activating 20 component carriers with both a centralized unit (CU) and distributed unit (DU) on a single NVIDIA GH200 server. SynaXG also demonstrated the first implementation of AI-RAN on FR2 (millimeter wave) bands, a crucial step for unlocking the full potential of 5G.

The Rise of the AI-RAN Alliance

The AI-RAN Alliance is playing a key role in driving innovation and standardization in this space. Over 20 demos built on NVIDIA platforms will be showcased at MWC, illustrating how AI is enhancing 5G performance and efficiency and enabling new edge AI applications. This year’s MWC will feature triple the number of AI-RAN innovations compared to last year, with 26 out of 33 Alliance demos utilizing NVIDIA AI Aerial and a software-defined architecture. According to Alex Choi, chair of the AI-RAN Alliance, this convergence around a software-defined, GPU-accelerated architecture is “boosting innovation, validating new concepts quickly and building the foundation for AI-native 6G, now.”

Innovations Beyond Core Performance

The benefits of AI-RAN extend beyond raw speed and latency. DeepSig is leveraging AI to improve how devices communicate with networks, allowing AI to learn a smarter signal format at both ends of the communication link. Early results on NVIDIA platforms suggest this approach could yield up to a 2x increase in throughput and improved spectral and energy efficiency. SUTD, NVIDIA, and partners are exploring how to distribute AI processing across devices, edge servers, and the cloud, optimizing for latency, privacy, and coverage. ZTouch Networks and partners have developed an AI-RAN orchestration blueprint that utilizes NVIDIA Multi-Instance GPU technology to dynamically allocate GPU resources between AI and RAN workloads, maximizing utilization and energy efficiency.

Northeastern University and SoftBank are demonstrating an AI switching solution that seamlessly transitions between AI and traditional algorithms for channel estimation, selecting the optimal processing method based on real-time conditions. This approach aims to improve stability and throughput while demonstrating the compatibility of AI with existing technologies.

Building an Ecosystem for AI-RAN

NVIDIA is fostering a growing ecosystem of partners to provide operators with a range of deployment options. Quanta Cloud Technology (QCT) is announcing commercial off-the-shelf AI-RAN products supporting NVIDIA ARC platforms and Nokia software. Supermicro is expanding support across the NVIDIA AI-RAN portfolio, including NVIDIA ARC-Pro and RTX PRO 6000 Blackwell Server Edition GPUs. WNC has introduced a new AI-optimized indoor and outdoor open radio unit integrated with NVIDIA AI Aerial Testbed and ARC platforms, supporting 5GA and 6G use cases. Eridan has launched a 4T4R O-RU integrated with NVIDIA AI Aerial, and a DU running on the NVIDIA DGX Spark desktop supercomputer. LITEON has completed integration of its radio units with NVIDIA AI Aerial and expanded collaboration with partners like Supermicro and SynaXG.

Looking Ahead: Open, Secure, and AI-Native 6G

NVIDIA’s recent State of AI in Telecom report indicates increasing industry investment in AI-native RAN and 6G, with 77% of respondents anticipating a faster deployment timeline for this new architecture. NVIDIA has also open-sourced NVIDIA Aerial CUDA-accelerated RAN libraries and joined the OCUDU (Open CU DU) Ecosystem Foundation, contributing to open-source RAN software development. These efforts are laying the groundwork for secure, open, and AI-native 6G systems, where intelligence is deeply integrated into the network fabric, enabling new applications in areas like autonomous mobility, as demonstrated by Capgemini’s Project ULTIMO, which utilizes NVIDIA Jetson Orin modules for processing sensor data in autonomous shuttles.