T-Mobile US Logs 16% Throughput Gain From AI Beamforming Trial

T-Mobile US recorded a 16% improvement in downlink user throughput by letting an AI model allocate radio resources on its 5G-Advanced network, with the largest gains appearing during peak hours. The trial, run with Ericsson on live commercial traffic at newly activated sites in New York City, also delivered about 50% more advanced beamforming using full-band SRS, 12% better spectral efficiency and a 10% increase in downlink data volume.

Chief network officer Ankur Kapoor outlined the results in a blog post. The operator disclosed neither the number of sites involved nor a baseline for the measurements, which limits comparison with conventional deployments.

The technology under test, AI-Managed Beamforming, targets sounding reference signals (SRS). Phones send these short transmissions so the network can measure the path between device and cell site and steer beams more precisely. SRS resources are finite, and operators have historically assigned them under fixed rules. T-Mobile instead trained a supervised-learning model on its own network data, running it on Ericsson's RAN compute and silicon to predict where demand rises and redirect resources accordingly.

Kapoor argued that home-grown training data gives the model an edge on T-Mobile's network that operators contributing no tuning data cannot match. The approach fits T-Mobile's spectrum position, he added, because beamforming performs particularly well in TDD bands, where uplink and downlink share the same frequencies, and T-Mobile holds a leading midband TDD footprint in the US.

What does this change for subscribers?

In practice, the gains land where congestion bites hardest: peak hours. A 16% downlink throughput improvement and 10% more downlink volume at busy sites mean more capacity from existing spectrum rather than new auctions or cell splits. No timetable for a wider rollout appeared in the post.

The New York results build on earlier joint work. In May the two companies reported close to 10% better spectral efficiency from a separate AI-native scheduler with link adaptation, also running on live traffic.

Kapoor framed the trial against rising demand. Ericsson's June 2026 Mobility Report put global mobile data traffic growth at 22% year-on-year, with video accounting for roughly 75% of traffic and 5G subscriptions above 3.1 billion.

Jaco du Plooy, SVP and head of the T-Mobile customer unit at Ericsson Americas, said the trial marked another step toward networks that respond more intelligently to demand, enabled by Ericsson's silicon. That phrasing is vendor positioning; the measured figures stand on their own, while claims of broader "intelligent" networks remain unquantified.

A proving ground for 6G

Kapoor linked the work to next-generation requirements. 6G networks, he said, will need to sense, learn and respond to support applications such as distributed AI, robotics, vision AI and other forms of physical AI. He described T-Mobile's nationwide 5G-Advanced network as its proving ground.

The regulatory and standards backdrop matters here. 5G-Advanced, the 3GPP Release 18 track, formalizes AI-driven RAN functions such as beam management and link adaptation, giving operators a standardized route to deploy models of this kind rather than bespoke overlays. Trials like T-Mobile's feed operator input into how far those functions scale commercially.

What comes next is commercialization at scale. Neither company set a date, but with a nationwide 5G-Advanced network in place and two positive live-traffic trials now logged, T-Mobile has a clear path to extend AI resource allocation beyond New York if the gains hold across broader load conditions.