SynaXG Proves the Art of the Possible with Fully Software-Defined AI-RAN on NVIDIA AI Aerial

SynaXG announced technical breakthroughs in AI-native radio access networks. The Singapore-based company demonstrated concurrent operation of 5G FR1, 5G FR2 and AI workloads on the NVIDIA AI-RAN platform with real-time, policy-driven GPU orchestration. The system achieved carrier-grade 5G FR1 performance on a single NVIDIA GH200 platform. It operated 20 cells of 100 MHz 5G NR spectrum. The setup delivered more than 36 Gbps of aggregated throughput with sub-10 millisecond latency while supporting up to 1,200 connected user equipment per cell. SynaXG reported a carrier-grade FR2 virtualized RAN implementation running concurrently with FR1 and AI workloads on a shared GPU platform. The setup delivered high throughput with end to end latency as low as 5 milliseconds. Real-time orchestration software dynamically allocates and switches Multi-Instance GPU partitions based on live RAN and AI key performance indicators without service interruption. The platform operated continuously 24 hours a day, seven days a week under sustained load. It is built on the NVIDIA AI Aerial platforms. SynaXG plans to extend the architecture to additional NVIDIA platforms, including DGX Spark. The results were showcased at the AI-RAN Alliance Working Group 2 demonstrations at the AI-RAN Alliance booth in Hall 2, 2E45 and at the SynaXG booth in Hall 5, 5F21 during Mobile World Congress 2026 in Barcelona. Xin Huang, CEO of SynaXG, said with the recent breakthroughs the company has demonstrated that AI-RAN can deliver carrier-grade FR1 and FR2 performance with continuous operation on shared NVIDIA AI infrastructure. Soma Velayutham, VP of AI and Telecoms at NVIDIA, said software-defined architecture is key to the next generation of wireless networks and that the benchmark performance proves operators and enterprises can achieve the flexibility and agility of cloud-native computing while maintaining carrier-grade throughput and performance per watt.