Speaking at the RCR Wireless News’s Telco AI Forum 2026, AT&T and Verizon executives described AI-based RAN deployments that are already producing measurable results
In sum – what to know
AI proves its value – AT&T and Verizon report measurable gains from AI-based RAN features, including improvements in throughput, spectral efficiency, and coverage.
Worthiness sets the gate – Both operators stress that AI must deliver meaningful performance benefits that justify additional compute, cost, and operational complexity.
Scaling changes the challenge – Moving beyond trials will require open architectures, automation, and disciplined lifecycle management across distributed networks.
AI is moving deeper into the radio access network, but for AT&T and Verizon, the next stage is less about proving that the technology works than determining where it creates enough value to justify deployment at scale.
Speaking at the RCR Wireless News’s Telco AI Forum 2026, Adam Loddeke, AVP RAN technology at AT&T, and Cindy Dong, associate director of RAN planning at Verizon, described deployments that are already producing measurable results while emphasizing the economic and operational constraints involved in expanding AI across the network.
Loddeke said AT&T is already using AI at scale, primarily through applications and rApps covering different stages of the network lifecycle. AI is used for planning and forecasting, deployment and configuration, as well as network optimization and lifecycle management.
The operator is also testing AI-native capabilities embedded directly into the RAN. Loddeke said AT&T has been trialing Ericsson’s AI-native link adaptation feature and has seen “significant improvements” in throughput and spectral efficiency.
Verizon is pursuing what Dong described as a dual RAN AI strategy. This combines centralized rApps operating in a non-real-time environment with local AI integrated into the gNodeB for real-time optimization. Dong said Verizon’s work with Ericsson has demonstrated improvements in downlink spectral efficiency from AI-based link adaptation, while another field trial produced coverage improvements.
For both operators, however, performance gains are only one part of the equation. Loddeke said the biggest current value from AI lies in optimization and automation, particularly where it can improve performance, lower costs, and reduce human involvement. He identified areas including beamforming, channel estimation, interference optimization, and traffic steering, alongside operational applications such as fault detection, root-cause analysis and closed-loop remediation.
Energy efficiency is another area of interest. Loddeke said AI could help reduce manual operations, accelerate issue resolution, and generate cost savings.
Dong similarly framed Verizon’s objective around achieving strong performance at network scale without significantly increasing costs. She said the operator sees particular potential in moving AI beyond centralized functions and into real-time processing at the gNodeB, including Layer 1 and Layer 2 functions.
That creates what Dong called an “AI worthiness” test. Any performance or user-experience improvement generated by AI needs to outweigh the additional computing cost and complexity introduced into the network. “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 principle also affects which problems the operators choose to address. Dong said the move from pilots to production depends on identifying the specific problem that is worth solving with AI. At Verizon, an AI feature must demonstrate a meaningful performance gain over the existing rule-based baseline.
Loddeke described a similar cost-benefit calculation. He noted that not every application needs to use more computationally intensive AI because additional processing carries a cost. Network AI must balance competing constraints such as capacity and performance, peak performance and energy savings, and speed and stability.
The complexity increases as AI becomes more distributed. Models may operate at the cell edge or closer to the network management and service management and orchestration layer, creating new requirements for model accuracy, update frequency, and lifecycle management.
For AT&T, Loddeke said operations teams need to understand how many models are deployed, how they are maintained, and how they affect network troubleshooting. Determining how quickly models should be updated and how they are applied will become critical as deployments expand.
Both executives also highlighted architecture and automation as foundations for scale. Loddeke said data provides the foundation, automation is required to scale through closed-loop execution, and architecture enables AI to operate across different parts of the network.
He emphasized the importance of an open, programmable platform that allows AT&T to work with third parties, vendors and its own teams rather than relying on a single source for AI development.
Dong agreed that architecture is fundamental but maintained that identifying the right problem remains the priority for Verizon. She said the operator is looking for areas where existing rule-based approaches have limitations and AI can provide additional gains without imposing disproportionate costs.
Over the next 12 to 18 months, Dong said Verizon’s priority is to maintain its long-term planning momentum while executing on selected AI areas and turning work developed over several years into tangible customer gains. Deployments will also need to pass the operator’s AI-worthiness test while maintaining controlled costs and improving user experience.
For Loddeke, the focus is on scaling new use cases, particularly at the gNodeB, while ensuring the architecture can support AI capabilities that emerge over the next several years.
