From demonstration to deployment

In a March 1, 2026 post, NVIDIA argues that software-defined AI-RAN has moved beyond controlled demos into live field milestones. Timed ahead of Mobile World Congress (March 2-5, Barcelona), the update combines operator deployment signals, benchmark data, and ecosystem expansion around NVIDIA AI-RAN platforms built with Nokia and other partners.

The central thesis is architectural: future wireless networks should run AI and RAN workloads on a shared, programmable compute foundation instead of siloed infrastructure. NVIDIA positions this as a practical path toward AI-native 6G.

Operator milestones and technical metrics

NVIDIA highlights three operator paths. T-Mobile U.S. demonstrated concurrent AI and RAN processing over the air in the 3.7GHz band using Nokia CUDA-accelerated RAN software on NVIDIA AI-RAN infrastructure. SoftBank reported a live AITRAS trial with 16-layer massive MIMO on fully software-defined 5G. Indosat Ooredoo Hutchison (IOH) moved from proof-of-concept to pre-commercial field validation and demonstrated an AI-powered 5G call workflow at MWC.

The most concrete performance numbers came from SynaXG, which ran software-defined AI-RAN with NVIDIA AI Aerial across 4G and 5G in both FR1 and FR2 bands, alongside agentic AI workloads, on a single NVIDIA GH200 server. NVIDIA says the configuration activated 20 component carriers and achieved 36 Gbps throughput with under 10 milliseconds latency.

Ecosystem and commercialization signal

NVIDIA says 26 of 33 AI-RAN Alliance demos at this year’s MWC are built on NVIDIA AI Aerial and a software-defined approach. It also points to expanding platform support from partners including QCT, Supermicro, WNC, Eridan, and LITEON, suggesting vendors are building toward commercially repeatable AI-RAN stacks rather than custom one-off deployments.

The post links this momentum to longer-horizon network transition. NVIDIA cites its State of AI in Telecom survey indicating 77% of respondents expect faster deployment of AI-native wireless architectures. It also notes that NVIDIA has open-sourced Aerial CUDA-accelerated RAN libraries and joined the Linux Foundation-hosted OCUDU ecosystem effort.

Why this matters

This is high impact because it reframes telecom AI from auxiliary optimization to core network design. If operators can reliably co-schedule RAN and AI workloads while preserving quality-of-service, AI-RAN becomes both a cost model and a new revenue platform for edge-native applications. In that scenario, 6G readiness is no longer only about future radio standards; it also depends on software orchestration, GPU resource governance, and production-grade multi-tenant network operations being proven now.

Source: NVIDIA Blog

#5g

NVIDIA Pushes Software-Defined AI-RAN Into Field Deployment Ahead of MWC 2026