AT&T, T-Mobile show AI in RAN with Ericsson sans GPUs

AT&T and T-Mobile US have conducted successful trials of artificial intelligence capabilities within their radio access networks using Ericsson’s AI-native link adaptation software. The North American carriers demonstrated that significant performance improvements can be achieved through machine learning without the requirement for dedicated graphics processing units. These tests focused on enhancing spectral efficiency and overall uplink throughput across their respective commercial infrastructures.

The partnership with the Swedish equipment manufacturer marks a shift in how network intelligence is deployed at the edge. By integrating AI-native link adaptation directly into the existing RAN compute architecture, the operators managed to optimise signal quality and data rates more effectively than traditional algorithms. This approach relies on standard central processing units, potentially lowering the total cost of ownership for 5G advanced features.

Ericsson’s software-based solution works by predicting the radio environment in real-time and adjusting transmission parameters accordingly. This allows the network to maintain high-speed connections even under challenging conditions or at the cell edge. During the testing phase, the participants reported measurable gains in data speeds and a noticeable reduction in latency for mobile users.

The move to avoid specialised hardware like GPUs is significant for the broader telecommunications industry. While high-end chips are often associated with complex AI workloads, these trials suggest that specific RAN functions can be handled by the general-purpose processors already present in modern base stations. This strategy could accelerate the adoption of intelligent networking by removing hardware bottlenecks and reducing power consumption.

Both AT&T and T-Mobile have expressed a commitment to integrating more automation into their networks to manage the increasing complexity of 5G traffic. By utilising AI for link adaptation, they aim to maximise the value of their existing spectrum holdings. The results of these trials are expected to inform the next phase of their network modernisation programmes.

The successful implementation of these features provides a blueprint for other global operators looking to upgrade their infrastructure using software-driven intelligence. As the industry moves toward 5G Advanced standards, the role of AI in managing radio resources is expected to expand further. The carriers will now evaluate the potential for a wider commercial rollout of this technology across their national footprints.