Nvidia and Partners Demonstrate Software‑Defined AI‑RAN as the Next Wireless Era

NVIDIA’s AI‑RAN platform is now proving its mettle in the field, with three major operators reporting live‑over‑the‑air milestones that move the technology beyond the lab. T‑Mobile U.S. ran concurrent AI and RAN workloads on a NVIDIA AI‑RAN system paired with Nokia’s CUDA‑accelerated AirScale massive MIMO radio in the 3.7 GHz band, delivering commercial‑grade video streaming, generative‑AI services and AI‑powered captioning on standard 5G devices (Kanika Atri, NVIDIA blog, Feb 28 2026). SoftBank’s AITRAS trial pushed the envelope further, achieving a 16‑layer massive MIMO configuration that runs entirely in software on NVIDIA hardware—an industry first that the company says “marks an important technical milestone toward AI‑RAN commercialization” (Atri). In Southeast Asia, Indosat Ooredoo Hutchison (IOH) moved from proof‑of‑concept to pre‑commercial validation, using Nokia’s vRAN stack on NVIDIA platforms to enable a cross‑border AI‑enhanced 5G call and real‑time control of a robotic dog, a demonstration slated for MWC Barcelona (Atri).

Performance benchmarks released by partner SynaXG underscore the raw capability of the NVIDIA stack. Running on a single NVIDIA GH200 server, the testbed combined 4G, sub‑6 GHz 5G (FR1) and millimeter‑wave 5G (FR2) bands with agentic AI workloads, activating 20 component carriers across both centralized (CU) and distributed (DU) units. The system delivered 36 Gbps of aggregate throughput while keeping latency under 10 ms, a “world’s first implementation of AI‑RAN on FR2 bands” according to the blog (Atri). SynaXG’s results demonstrate that AI‑RAN can simultaneously handle high‑speed data transport and compute‑intensive AI tasks without sacrificing carrier‑grade reliability, a claim echoed by the blog’s reference to “high‑speed, carrier‑grade performance … across multiple 5G spectrum bands” (Atri).

The momentum is translating into a growing ecosystem of demos and alliances that will converge on MWC. More than 20 AI‑RAN Alliance prototypes built on NVIDIA platforms are slated for exhibition, showcasing use cases ranging from edge‑AI video analytics to AI‑driven network slicing (Atri). Operators and vendors are co‑authoring open‑source software layers that abstract the underlying hardware, reinforcing the blog’s thesis that “software‑defined AI‑RAN is the only viable way to build future AI‑native wireless networks.” This collaborative push is intended to lay the groundwork for a “secure, open and AI‑native 6G system” that can be rolled out once the 5G‑to‑6G transition begins (Atri).

NVIDIA’s broader AI chip strategy provides the silicon backbone for these advances. Reuters reported that the company is developing a new AI chip based on its latest Blackwell architecture, aimed at markets such as China, and that CEO Jensen Huang believes the firm is “well positioned for the shift in AI” (Reuters, Feb 28 2026). While the Blackwell flagship faces a design‑flaw delay, the upcoming Rubin chip—integrating GPU, CPU and networking functions—promises to accelerate AI workloads across the telecom stack (Reuters). These hardware roadmaps dovetail with the AI‑RAN field trials, ensuring that the compute power needed for real‑time inference at the edge will be available as operators scale the technology.

Collectively, the live trials, benchmark data, and expanding partner ecosystem signal that AI‑RAN is moving from experimental proof to commercial reality. The field results from T‑Mobile, SoftBank and IOH demonstrate that software‑defined, AI‑native radio access can meet the reliability and latency thresholds demanded by carrier‑grade services, while SynaXG’s throughput and latency figures prove that the architecture can sustain multi‑band, high‑capacity traffic. With MWC poised to showcase over two dozen live demos, the industry appears ready to adopt a unified, software‑centric foundation that will underpin both the final phase of 5G and the forthcoming 6G era.