T-Mobile logs world's first AI-RAN call with Nokia, Nvidia

T-Mobile's twin announcements this week show it testing two different AI-RAN architectures at once. Nokia's approach relies on Nvidia's GPU-based Aerial RAN computer platform, while Ericsson uses its own silicon. The Nokia call, made in a live production network near T-Mobile's headquarters, is the first voice call of its kind according to Nokia, and comes about a year after the Finnish vendor partnered with Nvidia.

The Ericsson trial, run in T-Mobile's live 5G-Advanced network in New York City, delivered measurable gains during peak usage hours: 16% better downlink throughput, 12% better spectral efficiency and a 10% increase in downlink data volume. Omdia analyst Gabriel Brown called it a "wow moment" because the AI models run directly in the basestation and the gains came from software updates to deployed equipment. It was the second AI model T-Mobile deployed in Ericsson RAN this year, after an AI-native scheduler trial that showed 10% better spectral efficiency and a 15% increase in downlink throughput.

Brown noted that more AI models could be stacked, though it is not yet known how many can run simultaneously or what the combined gain would be. Nokia said its next field-trial phase will explore other spectrum bands and support AI inferencing workloads alongside RAN functions, suggesting the carrier is still mapping out how far these software-defined gains can go.