NVIDIA and partners demonstrate Software-Defined AI-RAN is the next wireless generation

The SynaXG setup activated 20 component carriers with both centralized units (CUs) and distributed units (DUs) on one platform, achieving a throughput of 36 Gbps and a latency of less than 10 ms. These breakthrough results highlight AI-RAN-based 5G performance and seamless orchestration between AI and RAN workloads.

Triple the pace of AI-RAN innovation

This year’s MWC tripled the number of AI-RAN innovations from last year, with 26 of the 33 AI-RAN Alliance demos built using NVIDIA AI Aerial and software-defined architecture.

These demos include:

• DeepSig is reinventing the way devices “talk” to networks by teaching AI to learn smarter signal formats at both ends of the link (the communication channel that connects two devices). The AI-native air interface uses neural techniques on devices and base stations to jointly learn how to best encode and decode signals, removing pilot overhead and adapting to site-specific channels. Initial results on the NVIDIA platform show up to approximately 2x higher throughput from the same spectrum, with improved spectral and energy efficiency.
• SUTD, NVIDIA, and partners will show how robots and self-driving cars can distribute their “thinking” across devices, the edge, and the cloud, enabling split reasoning from concept to implementation. This demo proves how AI-RAN can meet stringent latency, privacy, and coverage service level agreements and extend physical AI and vision language models through the network edge by determining where to execute each AI task in real time.
• zTouch Networks and partners have built an AI-RAN orchestration blueprint that shows operators how to securely share GPUs between AI and RAN workloads. Using NVIDIA multi-instance GPU technology, Blueprint controls resources in real-time to maximize GPU utilization and improve energy management while ensuring RAN quality of service. This is an important step towards commercially enabling multi-tenant AI-RAN solutions, allowing carriers to turn GPU capacity into revenue.
• Northeastern University and SoftBank will demonstrate an AI switching solution for NVIDIA AI Aerial that switches between AI and traditional algorithms for channel estimation in microseconds. This ensures that the best processing solution is always selected in real time depending on the conditions, improving stability and throughput, while proving that AI can coexist with traditional approaches.

“AI-RAN is emerging as the unified architecture for future wireless networks,” said Alex Choi, Chairman of the AI-RAN Alliance. “Operators, vendors, and researchers are now collaborating around a software-defined, GPU-accelerated architecture to accelerate innovation, quickly validate new concepts, and build the foundation for AI-native 6G.”

As intelligence moves into the physical world, autonomous systems such as robots and cars rely on AI-RAN networks to see, sense, reason, and act.

Capgemini is working on Project ULTIMO, an initiative funded by Horizon Europe to demonstrate how AI-RAN can support large-scale autonomous mobility services across European cities. Autonomous shuttles equipped with NVIDIA Jetson Orin modules process sensor data locally, and selected video and telemetry streams are sent over 5G to agent AI applications on NVIDIA AI-RAN servers. These workloads process field understanding, incident and safety detection, and accessibility insights at scale, while mission-critical 5G receives preferential access to GPU resources.

A growing ecosystem

An ecosystem of partners is growing around the NVIDIA-powered AI-RAN platform, giving carriers a wide range of deployment solutions to choose from. The NVIDIA Aerial RAN Computer (ARC) platform leverages NVIDIA Grace CPUs and various GPUs to provide a high-performance, energy-efficient computing foundation for AI-native RAN infrastructure.

• Quanta Cloud Technology (QCT) announces a commercial off-the-shelf AI-RAN product supporting the NVIDIA ARC platform and Nokia software, providing carriers with standardized building blocks for AI-RAN.
• Supermicro is expanding support across the entire NVIDIA AI-RAN portfolio, including NVIDIA ARC-Pro and NVIDIA RTX 6000-based configurations, and ARC-Compact systems with Nokia software.
• WNC has introduced a new AI-optimized indoor and outdoor open radio unit that integrates with the NVIDIA AI aviation testbed and NVIDIA ARC platform and supports 5GA and 6G use cases.
• Eridan launched the 4T4R O-RU along with the 2T2R O-RU integrated with NVIDIA AI Aerial and DU running on NVIDIA DGX Spark desktop supercomputers. It combines spectrally efficient radios and GPU-based baseband processing to create a powerful and portable outdoor base station.
• LITEON has completed the integration of sub-6 GHz and mmWave radio units with NVIDIA AI Aerial and expanded collaboration with ecosystem partners such as Supermicro and SynaXG to accelerate the commercialization of AI-RAN.

Laying the foundation for open, secure, AI-native 6G

NVIDIA’s latest State of AI in Telecom report shows that the industry is ramping up investments in AI-native RAN and 6G, with a massive intercept coming ahead of traditional 6G deployment cycles, with 77% of respondents expecting significantly faster time-to-deployment of this new AI-native wireless network architecture.

This latest advancement in software-defined AI-RAN is poised for secure, open, AI-native 6G systems.

NVIDIA has already open sourced the NVIDIA Aerial CUDA-accelerated RAN library, accelerating the pace of AI-RAN innovation. NVIDIA also participates in the Linux Foundation’s OCUDU (Open CU DU) Ecosystem Foundation, contributing to the development of open source RAN software to accelerate research and commercialization of next-generation wireless networks.

Meet our partners at NVIDIA to learn more. mobile world congress. Explore key insights from Current state of AI in communications investigation.