T-Mobile has succeeded in what it called the world’s first artificial intelligence radio access network (AI-RAN) call over a commercial 5G base station with Nokia and Nvidia.

On Oct. 27, local time, T-Mobile and TechRadar and other foreign media reported the call took place at a commercial base station in Bellevue, Washington state. It is significant in that the call connected T-Mobile’s live 5G standalone (SA) core network with a commercial smartphone.

The demonstration used Nokia’s Massive MIMO radio equipment and 5G anyRAN software, along with a server based on Nvidia’s Aerial RAN Computer (ARC) platform. The key was shifting the RAN processing structure from dedicated-chip centric designs to GPU-accelerated hardware. Nokia’s CUDA-based radio software runs on Nvidia GPUs to optimise wireless connections and use spare computing resources for other AI tasks.

The main goal of AI-RAN is to improve spectrum efficiency by transmitting more data in the same frequency band. Nokia set a goal of doubling spectrum efficiency through AI-RAN. It said a platform unveiled in July 2026 had achieved more than a 20 percent efficiency improvement so far, and set targets of 50 percent in 2027 and more than 100 percent in 2028. It did not disclose specific measurement conditions.

In the call, the three companies did not disclose actual data throughput, latency or spectrum-efficiency improvement figures. The demonstration therefore proved that Nokia’s radio software works on Nvidia hardware in a live commercial network, but it did not confirm how much AI improves call quality.

There are cases of performance improvements for related technologies. Deutsche Telekom said initial trials in Bonn and Oulu found uplink throughput typically rose 10 to 15 percent through AI-based channel estimation. It said those were laboratory test results, and field verification has not yet been carried out.

T-Mobile is also testing AI-RAN technology that does not use Nvidia GPUs. A day before announcing this demonstration, it said tests in New York of Ericsson’s in-house semiconductor-based AI beamforming technology improved downlink user throughput by 16 percent and spectrum efficiency by 12 percent during peak congestion periods. It previously said in May that spectrum efficiency improved by about 10 percent through a separate AI scheduler that operates without GPUs. It said those figures were all results disclosed by companies, and T-Mobile did not reveal the number of base stations involved in the tests.

Gabriel Brown (가브리엘 브라운), a principal analyst at Omdia, described the achievement as an "astonishing moment." He highlighted that performance can be improved by applying software updates to existing field equipment.

T-Mobile set up an AI-RAN Innovation Center in 2024 with Nvidia, Ericsson and Nokia, and is verifying technologies from both vendors. Ankur Kapoor (안쿠르 카푸어), T-Mobile’s chief network officer (CNO), said AI-RAN is "not a product but a continuum" and that it is developing related software with the two companies.

Commercialisation and cost validation remain tasks. Nokia’s AI-RAN product is to begin pilot deployments around the end of 2026, with commercial supply targeted for 2027. John Saw (존 소), T-Mobile’s chief technology officer (CTO), said it will gauge total cost of ownership (TCO) through field tests. He also showed a position that Nokia’s spectrum-efficiency improvement targets must be proven with actual performance.

Even if the technology is commercialised, users will not immediately feel changes in call audio quality. The key is whether it can increase network capacity in congested areas. The demonstration is also significant for Nvidia. The more GPUs are installed in base stations, the more new revenue opportunities arise for Nvidia. Jensen Huang, Nvidia’s chief executive officer (CEO), said if communications networks have so far been built for conversations, they will in the future also carry "the world’s intelligence."