T-Mobile and Ericsson report near-10% spectrum efficiency gains and up to 15% higher downlink throughput from AI-native RAN trials in the U.S.
T-Mobile is taking its AI-native RAN ambitions from the lab to live networks, and the early results from trials with Ericsson are already drawing attention for what they could mean for everyday mobile performance.
The companies say that an “AI-native Scheduler with Link Adaptation” tested on T-Mobile’s 5G Advanced network lifted spectrum efficiency by nearly 10% in U.S.. trials.. They also report downlink throughput improvements of up to 15% compared with legacy. rule-based scheduling methods—an outcome they describe as both meaningful for network efficiency and relevant for customer experience.
Trials began in early second quarter 2025 and expanded across multiple U.S.. markets.. The rollout covered cities including Los Angeles. New York. New Jersey. and Salt Lake City. allowing the software to be evaluated across different real-world operating conditions rather than a single controlled area.
T-Mobile’s leadership ties the testing push to its broader push for RAN innovation following a major deployment milestone.. The report states that the company has already positioned itself as the first U.S.. operator to deploy 5G Advanced nationwide in 2025. and it is now aiming to commercialize the AI-native scheduling technology during the third quarter of this year.
The trial findings center on how the scheduler works.. Instead of relying on fixed rules. the AI-native scheduler is designed to optimize how the network allocates resources as conditions change.. That distinction matters because wireless networks can fluctuate rapidly with demand and radio conditions. meaning software that can adapt in real time is often viewed as a route to more consistent performance.
Ericsson’s role in the trial is framed around ensuring that the optimization holds up when networks face tougher conditions.. The report says Ericsson’s AI-native Scheduler with Link Adaptation is intended to maintain reliable performance even in high-demand environments where radio frequency conditions may be less favorable.
For users, the companies link those engineering goals to practical outcomes. Ericsson says the approach can translate into smoother streaming, more responsive gaming, and uninterrupted video calls, particularly during peak usage when networks are under the most strain.
Ericsson also positions AI as a core element of its longer-term network strategy. The report states that its view is to embed intelligence directly into RAN software so programmable networks can deliver real-time performance improvements while helping operators maximize the value of their spectrum.
T-Mobile’s collaboration web extends beyond Ericsson in this effort. The report notes that both Ericsson and Nokia are part of T-Mobile’s collaboration with Nvidia, reflecting a broader ecosystem approach to AI-enabled networking rather than a single-vendor pathway.
The work traces back to T-Mobile’s earlier planning. In September 2024, T-Mobile announced an AI-RAN Innovation Center at its headquarters in Bellevue, Washington, where it said it would collaborate with Nvidia, Ericsson, and Nokia.
That same announcement also highlighted that the organizations involved are founding members of the AI-RAN Alliance. an industry initiative that began at Mobile World Congress in Barcelona in February 2024.. The report suggests the alliance is part of a wider push to advance AI in radio access networks across the telecom sector.
If T-Mobile moves from trial into commercialization as targeted. it could become one of the clearest early signals of how “AI-native” RAN features will translate into measurable network gains at scale.. Near-10% spectrum efficiency improvements. along with the reported throughput lift. are the kind of results operators typically watch closely because they can affect how effectively networks handle more traffic without proportionally increasing capacity.
The emphasis on performance during peak periods is also likely to shape customer expectations. When networks face congestion, the scheduling layer becomes crucial—so software that adapts dynamically may reduce the sharp variability that users often notice during busy hours.
Meanwhile, the multi-market geography of the trials underscores another practical point: real networks differ in demand patterns and radio environments. Expanding across major metro areas suggests the companies are looking for consistency, not just a one-off win in a single location.
The push also reflects how RAN modernization is increasingly tied to partnerships in the AI hardware and software stack.. With Nvidia. Ericsson. and Nokia named in the collaboration. the broader story is not only whether scheduling gets smarter. but also whether the ecosystem around AI-native networking can support deployment timelines like the third-quarter commercialization target mentioned in the report.
As T-Mobile prepares to bring the technology to market. the key question will be whether the reported gains hold up when traffic patterns. device mix. and day-to-day network variables differ from controlled trial conditions.. For now. the companies say their AI-native scheduler with link adaptation is already showing performance gains against legacy approaches—an early foothold for what may become a more AI-driven future for RAN operations.
T-Mobile Ericsson AI-native RAN 5G Advanced spectrum efficiency downlink throughput Nvidia collaboration