Operators are moving artificial intelligence deeper into the radio access network. Yet the next phase looks measured, not experimental. AT&T and Verizon now report real performance gains from AI-based RAN tools.
At RCR Wireless News’s Telco AI Forum 2026, executives shared early field results. Adam Loddeke, AVP RAN technology at AT&T, outlined current AI use cases. Cindy Dong, associate director of RAN planning at Verizon, described a dual AI strategy.
AT&T already uses AI across planning, forecasting, deployment, and network optimization. It also applies rApps across parts of the network lifecycle. These apps support automation and help engineers manage growing network complexity.
The company is also testing AI-native features inside the RAN itself. Loddeke said AT&T trials with Ericsson delivered “significant improvements” in throughput and spectral efficiency. These gains matter because spectrum remains costly and limited.
Meanwhile, Verizon is combining two approaches. Centralized rApps handle slower network intelligence tasks. Local AI inside the gNodeB supports faster decisions near the radio site.
Dong said Verizon saw better downlink spectral efficiency through AI-based link adaptation. Another field trial also improved coverage. That points to stronger user experiences without always adding new sites.
However, both operators stressed a practical message. AI must earn its place in the network. Extra computing, power needs, and software management all add cost.
Dong called this the “AI worthiness” test. “We don’t want to just deploy and scale the AI just for sake of doing AI,” Dong said. “We want to use this one to really benefit our customer.”
That test sets a high bar for deployment. Verizon wants AI features to beat current rule-based methods. Rule-based systems follow fixed instructions, while AI can adjust from patterns.
AT&T sees major value in optimization and automation. Loddeke pointed to beamforming, interference handling, and traffic steering. These functions help direct signals and manage network load more efficiently.
AI can also support fault detection and root-cause analysis. In simple terms, it can find problems faster. It may also trigger fixes without waiting for manual action.
Energy efficiency remains another major target. AI could reduce wasted power during low-traffic periods. It could also help operators balance performance and savings.
Still, distributed AI brings new operational questions. Operators must track many models across many network points. They also need clear update schedules and testing methods.
This becomes harder as models move closer to the radio edge. A poor model update could affect service quality. So lifecycle management becomes as important as the AI model itself.
Architecture will decide how far these deployments can scale. AT&T wants open and programmable platforms. That approach allows work with vendors, internal teams, and third parties.
Looking ahead, Verizon aims to turn selected AI work into customer gains. AT&T plans to scale new gNodeB use cases. For both, AI in the RAN must prove measurable value before broad rollout.
