Nokia, T-Mobile, and NVIDIA completed the call on October 21 at a commercial cell site in Bellevue, Washington. The site belongs to T-Mobile’s live production network. The radio was a Nokia Massive Multiple-Input Multiple-Output unit. The software was Nokia’s Fifth Generation anyRAN. The servers ran on the NVIDIA Aerial RAN Computer platform, which the companies call ARC. The site stayed connected to a commercial Fifth Generation core and ordinary commercial phones. Nokia’s Chief Technology and AI Officer Pallavi Mahajan published the account on the Nokia blog the same day. NVIDIA’s newsroom confirmed the milestone (Nokia, 2026; NVIDIA, 2026).
Nokia describes the event as the first software-defined Radio Access Network call that used accelerated computing on a live network. The company says the next field trials will test additional artificial intelligence algorithms for the radio and will run artificial intelligence inference workloads beside the radio functions on the same accelerated platform. Nokia has stated that the approach is on a path toward twice the spectral efficiency from spectrum operators already own (Nokia, 2026).
I wrote about Nokia’s anyRAN functional tests with T-Mobile, Indosat, and SoftBank in March. The Bellevue call is the first time that combination ran on a production cell site.
The AI-RAN Alliance exists. The hardware paths do not match.
The AI-RAN Alliance launched at Mobile World Congress in Barcelona in February 2024. Founding members included Nokia, Ericsson, NVIDIA, T-Mobile, SoftBank, Samsung, Amazon Web Services, Microsoft, and Arm. The Alliance works on artificial intelligence for the radio, artificial intelligence running with the radio on shared compute, and artificial intelligence services offered from the network edge. Membership has grown past 100 organizations (Business Wire, 2024; Computer Weekly, 2025).
Nokia’s path places its anyRAN software on NVIDIA’s graphics processing unit platform. The company says this supplies a software upgrade path toward Sixth Generation networks and Open Radio Access Network compliance, with three hardware options that include a plug-in card for existing AirScale radios. Commercial availability is targeted for 2027 after pilots that begin later this year. NVIDIA says nearly 20 operators have announced AI-RAN work that uses Nokia anyRAN software and the Aerial RAN Computer (Nokia, 2026; NVIDIA, 2026).
Ericsson runs its AI in RAN software on the Ericsson silicon and radios already deployed. The company says the first features do not require new graphics processing unit hardware. Ericsson reports up to 10 percent spectral efficiency gains and up to 20 percent higher downlink throughput in commercial deployments and trials. Features such as AI-native link adaptation and AI-managed beamforming are in live networks. Ericsson is a founding member of the AI-RAN Alliance and contributes to the Third Generation Partnership Project and the Open Radio Access Network Alliance (Ericsson, 2026).
T-Mobile hosts the AI-RAN Innovation Center in Bellevue that it announced in September 2024 with Nokia, Ericsson, and NVIDIA. The live call this week is the Nokia-NVIDIA milestone. Separate Ericsson AI-managed beamforming trials on T-Mobile’s Fifth Generation Advanced network were reported on October 20, with measured gains in beamforming use and downlink throughput (T-Mobile, 2024; T-Mobile, 2026).
Who pays for the graphics processing units at every cell site
Artificial intelligence applications need low latency at the edge. A robot or a vision agent cannot wait for a distant data center when the model must decide in milliseconds. Shared accelerated computing at the cell site is one way to place that capacity near the radio.
The operator who chooses the Nokia-NVIDIA path must fund graphics processing units at sites that today hold purpose-built radio hardware. Those units can run radio functions and external artificial intelligence workloads on the same box. The revenue from the external workloads has to cover the power, cooling, and capital cost of the units, or the radio gains alone must justify the spend. Ericsson’s path embeds the inference in the silicon already paid for. The spectral gains are smaller on the published numbers, and the hardware refresh cycle is the one the operator already runs.
Both approaches can shorten the round-trip time for edge artificial intelligence. They do not share one common interface or one set of performance benchmarks an operator can use to compare them side by side.
Operator CIO viability check
Ask your radio vendor for the contract language that locks the software upgrade path for the next three releases. Demand a clause that names the compute platform the radio software will continue to run on, and a second clause that states whether artificial intelligence inference workloads can move to another vendor’s accelerated hardware without a forklift. Then run the arithmetic on whether external artificial intelligence jobs at the cell site will cover the power and capital of graphics processing units, or whether the spectral gain on existing silicon is enough.
Sources
- Pallavi Mahajan, “From the lab to the live network: Nokia, T-Mobile and NVIDIA complete world’s first AI-RAN call,” Nokia, 21 Oct. 2026, nokia.com.
- “NVIDIA, Nokia and Leading Telecom Operators Turn Networks Into Distributed Intelligence Platforms,” NVIDIA, 21 Oct. 2026, nvidia.com.
- Nokia, “Nokia defines the next era of radio with the industry’s first AI-native RAN platform,” 15 July 2026, nokia.com.
- Ericsson, “The RAN gets smarter: Ericsson puts AI where it matters,” 11 June 2026, ericsson.com.
- Business Wire, “Industry Leaders in AI and Wireless Form AI-RAN Alliance,” 26 Feb. 2024, businesswire.com.
- “AI-RAN Alliance draws more than 100 members in first year,” Computer Weekly, 10 July 2025, computerweekly.com.
- T-Mobile, “T-Mobile Announces Technology Partnership with NVIDIA, Ericsson and Nokia,” 18 Sept. 2024, t-mobile.com.
- T-Mobile, “5G Advanced in Action, Better Experiences: T-Mobile Puts AI to Work for Customers,” 20 Oct. 2026, t-mobile.com.
- Shashi Bellamkonda, “Nokia’s Strategic Pivot: From Connecting Devices to Connecting Intelligence,” shashi.co, Mar. 2026.
