LITEON and NVIDIA Partner on AI-RAN
What happened
At MWC 2026, LITEON Technology announced a partnership to accelerate AI-RAN commercialization. The collaboration integrates LITEON's open radio units with NVIDIA's AI Aerial platform, aiming to create more scalable, AI-native radio access network architectures.
Why it matters
The NVIDIA Aerial platform underpinning this partnership is a software-defined, GPU-accelerated framework for building 5G and 6G networks. For developers, NVIDIA provides a suite of open-source tools available on GitHub, including the Aerial CUDA-Accelerated RAN, which offers libraries for Layer 1 and Layer 2 signal processing. This allows for a modular and programmable approach to building the radio access network stack. A key component for developers is the NVIDIA Aerial Framework, which can generate high-performance, CUDA-accelerated 5G/6G pipelines from Python or MATLAB code. This is complemented by pyAerial, a Python API that enables the integration of AI and machine learning models, using frameworks like TensorFlow or PyTorch, directly into the physical layer of the network for real-time inference and optimization. LITEON's contribution is the carrier-grade Open Radio Unit (O-RU) which handles the radio frequency transmission and reception. These units are compliant with the O-RAN 7.2x fronthaul interface, ensuring standardized and reliable communication between the radio hardware and NVIDIA's GPU-based baseband processing. This open architecture is designed to reduce integration complexity and accelerate deployment. The collaboration extends beyond just LITEON and NVIDIA, involving partners like SynaXG and Supermicro to create a unified architecture for both RAN and AI workloads. In a demonstration by SynaXG, this integrated AI-RAN platform, running on a single NVIDIA GH200 server, achieved a throughput of 36 Gbps with a latency of under 10 milliseconds. This software-defined approach allows for dynamic resource allocation between network functions and AI applications. For instance, T-Mobile has demonstrated running concurrent AI tasks, such as generative AI and video captioning, alongside 5G RAN processing on the same platform. This shared infrastructure model aims to maximize GPU utilization and create new revenue opportunities for operators through edge AI services. For hands-on development and research, NVIDIA offers the Sionna Research Kit, powered by the Jetson platform, and the Aerial Omniverse Digital Twin. The digital twin platform allows for the simulation and testing of AI-RAN algorithms in physically accurate, large-scale virtual environments before deploying them on real-world hardware like the LITEON O-RUs.
Key numbers
- At MWC 2026, LITEON Technology announced a partnership to accelerate AI-RAN commercialization.
- The NVIDIA Aerial platform underpinning this partnership is a software-defined, GPU-accelerated framework for building 5G and 6G networks.
- For developers, NVIDIA provides a suite of open-source tools available on GitHub, including the Aerial CUDA-Accelerated RAN, which offers libraries for Layer 1 and Layer 2 signal processing.
- A key component for developers is the NVIDIA Aerial Framework, which can generate high-performance, CUDA-accelerated 5G/6G pipelines from Python or MATLAB code.
What happens next
- This shared infrastructure model aims to maximize GPU utilization and create new revenue opportunities for operators through edge AI services.
Quick answers
What happened in LITEON and NVIDIA Partner on AI-RAN?
At MWC 2026, LITEON Technology announced a partnership to accelerate AI-RAN commercialization. The collaboration integrates LITEON's open radio units with NVIDIA's AI Aerial platform, aiming to create more scalable, AI-native radio access network architectures.
Why does LITEON and NVIDIA Partner on AI-RAN matter?
The NVIDIA Aerial platform underpinning this partnership is a software-defined, GPU-accelerated framework for building 5G and 6G networks. For developers, NVIDIA provides a suite of open-source tools available on GitHub, including the Aerial CUDA-Accelerated RAN, which offers libraries for Layer 1 and Layer 2 signal processing. This allows for a modular and programmable approach to building the radio access network stack. A key component for developers is the NVIDIA Aerial Framework, which can generate high-performance, CUDA-accelerated 5G/6G pipelines from Python or MATLAB code. This is complemented by pyAerial, a Python API that enables the integration of AI and machine learning models, using frameworks like TensorFlow or PyTorch, directly into the physical layer of the network for real-time inference and optimization. LITEON's contribution is the carrier-grade Open Radio Unit (O-RU) which handles the radio frequency transmission and reception. These units are compliant with the O-RAN 7.2x fronthaul interface, ensuring standardized and reliable communication between the radio hardware and NVIDIA's GPU-based baseband processing. This open architecture is designed to reduce integration complexity and accelerate deployment. The collaboration extends beyond just LITEON and NVIDIA, involving partners like SynaXG and Supermicro to create a unified architecture for both RAN and AI workloads. In a demonstration by SynaXG, this integrated AI-RAN platform, running on a single NVIDIA GH200 server, achieved a throughput of 36 Gbps with a latency of under 10 milliseconds. This software-defined approach allows for dynamic resource allocation between network functions and AI applications. For instance, T-Mobile has demonstrated running concurrent AI tasks, such as generative AI and video captioning, alongside 5G RAN processing on the same platform. This shared infrastructure model aims to maximize GPU utilization and create new revenue opportunities for operators through edge AI services. For hands-on development and research, NVIDIA offers the Sionna Research Kit, powered by the Jetson platform, and the Aerial Omniverse Digital Twin. The digital twin platform allows for the simulation and testing of AI-RAN algorithms in physically accurate, large-scale virtual environments before deploying them on real-world hardware like the LITEON O-RUs.
