- Nvidia emphasized its work with Nokia and European operators during this week’s GTC Berlin event
- Among the announcements: Nokia completed what’s being dubbed a world first – an AI-RAN call on T-Mobile’s network in Bellevue, Washington
- According to T-Mobile’s Ankur Kapoor, the U.S. operator is advancing AI-RAN through two complementary approaches, one via Nokia and another with Ericsson
It’s been a week for Nokia and AI-RAN, thanks to Nvidia’s grand shindig at GTC Berlin and a string of announcements that included Elisa, BT, Deutsche Telekom and Orange.
The big kahuna for AI-RAN watchers in the U.S.: T-Mobile, which conducted the world’s first AI-RAN call at a commercial cell site in Bellevue, Washington, on Saturday, October 17.
The accomplishment was significant enough to merit a mention by Nvidia CEO Jensen Huang on the GTC stage on Wednesday.
“Just this last week, after working on it for nearly eight years, Nvidia made the first commercial phone call on the T-Mobile network with Nokia,” he said. “Now, CUDA makes phones calls, the traditional way. Then once it can do that, then we can layer AI on top of it. And so the revolution is going to begin.”
For AI-RAN aficionados, Nokia provided a few technical details, explaining that the call was made over T-Mobile’s live production network using Nokia Massive MIMO radio, Nokia 5G anyRAN software and server infrastructure built on the Nvidia Aerial RAN Computer platform. The site was connected to a commercial 5G core and commercial end-user devices.
Nokia Chief Technology and AI Officer Pallavi Mahajan hinted at some of the work that went on behind the scenes, noting that it took months of engineering and close collaboration across the companies, including design and staging through site preparation, integration, configuration, testing, optimization and final system set-up.
“This is not just the world’s first AI-RAN call, it is the world's first software-defined RAN call using accelerated computing. A moment only spoken of, never realized in the last four decades of wireless evolution. A moment when RAN is at parity with the cloud and software worlds, benefitting from continuous CI/CD innovation and evolution,” she said in a blog post.
That sounds quite dramatic.
Nvidia’s take on T-Mobile’s call
Huang’s comments about Nvidia’s work the past eight years refer to the development they’ve been undertaking in terms of software-defined RAN, according to Ronnie Vasishta, SVP Telecom at Nvidia, who discussed Nvidia’s vast telco-related news in a press briefing today.
Regarding the significance of the call with T-Mobile for Nvidia, Vasishta said it shows that “you can achieve the right TCO and KPIs expected of a commercial network” on the Nvidia Aerial RAN computer with Nokia software.
“It’s the first step as we move through the commercialization” with Nokia and its 19 or so telco partners around the world that are in various phases of engagement, he said, noting that carriers like Elisa in Finland are already talking about how they’ll deploy it live commercial networks.
T-Mobile’s work with AI-RAN
Nokia and Nvidia are drumming up a lot of accolades for their work, for good reason.
But obviously, they aren’t the only game in town. Earlier this week, T-Mobile touted how its work with Ericsson put AI-powered beamforming to the test in New York City, achieving a 16% improvement in download throughput during a network trial.
Fierce was curious how T-Mobile’s work with Nokia compares with what it’s doing with Ericsson. T-Mobile uses both Ericsson and Nokia infrastructure in its U.S. network.
With Nokia, it’s exploring a GPU-based AI-RAN architecture that brings together RAN processing and AI workloads on a flexible compute platform, “creating opportunities for future network improvements in spectral efficiency and edge AI applications,” said T-Mobile Chief Network Officer Ankur Kapoor.
“This is an important step toward the next generation of AI-powered wireless networks. Beyond demonstrating a new approach to RAN compute, this milestone lays the foundation for future AI-driven improvements in spectral efficiency while advancing our vision of a multipurpose network platform that supports communications, sensing and edge AI,” he told Fierce via email.
Not one to play favorites with vendors (as far as we’re aware), Kapoor said what makes this approach particularly interesting is they’re advancing AI-RAN through two complementary approaches:
- Nokia: Exploring a GPU-based AI-RAN architecture that brings together RAN processing and AI workloads on a flexible compute platform, creating opportunities for future network optimization and new AI applications.
- Ericsson: Already demonstrating AI-driven spectral efficiency improvements using existing purpose-built RAN hardware in T-Mobile's commercial network, including a 10% gain from AI-based link adaptation and a 12% gain from AI-powered beamforming in separate demonstrations.
“From T-Mobile's perspective, there isn't a single path to AI-RAN. Different architectures and implementation approaches can help us achieve the same long-term vision: a more intelligent, efficient and flexible network that can support both connectivity and emerging AI applications,” Kapoor said.
Read more stories about AI-RAN on Fierce Network:
Nokia and Nvidia keep AI-RAN momentum going as they edge closer to commercial reality
Nokia’s AI-RAN pitch gets reality check in new study
T-Mobile cites AI-RAN gains with Ericsson in NYC
Nvidia GTC: T-Mobile and Nvidia push physical AI to the network edge