T-Mobile completes claimed world-first AI-RAN call on 5G

T-Mobile completed what it called the world's first artificial intelligence radio access network call on Oct. 27 at a commercial 5G base station in Bellevue, Washington. The live 5G standalone core connected with a commercial smartphone. Nokia supplied the radio system, while Nvidia supplied the GPU computing platform.

T-Mobile, TechRadar and other foreign media reported the result on Oct. 27 local time. The radio was not confined to a test bench: Nokia's CUDA-based software ran on Nvidia GPUs inside the live network.

T-Mobile chief network officer Ankur Kapoor said AI-RAN is "not a product but a continuum." For technology still waiting on pilots and supply, that distinction keeps expectations in their proper place.

What did the Bellevue demonstration prove?

The setup used Nokia's Massive MIMO radio equipment and 5G anyRAN software, plus a server based on Nvidia's Aerial RAN Computer (ARC) platform. It moved RAN processing away from dedicated-chip designs toward GPU-accelerated hardware, while Nokia's software optimized wireless connections and left spare computing capacity for other AI workloads.

This established that Nokia's radio software can run on Nvidia hardware in a live commercial network. It did not establish how much AI improves call quality, because the companies disclosed no actual throughput, latency or spectrum-efficiency figures for the call.

Nokia's stated aim is to double spectrum efficiency, meaning more data within the same frequency band. The company said a platform unveiled in July 2026 had achieved:

• More than 20% improvement so far.
• A 50% target for 2027.
• More than 100% targeted for 2028.

Nokia did not disclose the measurement conditions. Without comparable field data, the percentages remain vendor claims rather than demonstrated network results.

What do other trials show?

Deutsche Telekom offers a related result from a different setting. Initial trials in Bonn and Oulu typically raised uplink throughput by 10% to 15% through AI-based channel estimation, but they were laboratory tests. The operator has not yet verified those results in the field.

A day before the Bellevue announcement, T-Mobile disclosed New York tests of Ericsson's in-house, semiconductor-based AI beamforming. During peak congestion, the operator reported 16% higher downlink user throughput and 12% better spectrum efficiency.

In May, T-Mobile said a separate AI scheduler operating without GPUs improved spectrum efficiency by about 10%. It did not reveal how many base stations joined these tests, leaving the percentages without deployment scale.

Can AI-RAN become a product?

T-Mobile set up its AI-RAN Innovation Center in 2024 with Nvidia, Ericsson and Nokia. The center is verifying technologies from both radio vendors, while T-Mobile is developing related software with the two companies.

Omdia principal analyst Gabriel Brown found the achievement astonishing and highlighted that software updates can improve performance on field equipment already deployed. That makes the software route attractive, but an upgrade still needs a price, a deployment plan and measured savings.

Nokia plans pilot deployments around the end of 2026 and commercial supply in 2027. T-Mobile chief technology officer John Saw said the carrier will assess total cost of ownership through field tests and wants Nokia's spectrum-efficiency targets proven against actual performance.

Commercialization and cost validation remain unfinished business. Even after commercial supply starts, users will not immediately hear better call audio; the practical prize is more network capacity in congested areas.

For Nvidia, each additional AI-RAN base station creates another GPU installation and another potential revenue opportunity. Nvidia chief executive officer Jensen Huang said: "If communications networks have so far been built for conversations, they will in the future also carry the world's intelligence."

The next checkpoints are Nokia's end-2026 pilots, T-Mobile's total-cost field tests and 2027 commercial supply. If measured gains survive those tests, AI-RAN will matter first in crowded cells. If they do not, Bellevue will remain a compatibility milestone rather than a capacity upgrade.