A recent Cloud RAN trial has demonstrated how artificial intelligence can significantly improve mobile network efficiency, offering operators a practical route to better resource management and reduced operational costs.
AT&T, working with Ericsson and Intel, carried out an evaluation that measured the performance of machine learning models across different radio frequencies. The experiment ran on a cloud-native architecture powered by Intel processors. By using commercial off-the-shelf hardware instead of proprietary radio equipment, operators can lower total cost of ownership and avoid vendor lock-in while accelerating feature rollout.
Telecom providers are increasingly testing AI-driven techniques to extract more value from radio networks. One such approach—AI-native link adaptation—monitors channel conditions and interference in real time and selects optimal transmission rates dynamically rather than relying on fixed rules.
In the trial, this adaptive method produced throughput improvements of up to 20 percent and raised spectral efficiency, creating a concrete benchmark for how AI can enhance radio access network performance.
Overcoming legacy infrastructure barriers
Implementing these AI capabilities requires a shift away from tightly integrated legacy systems. The trial ran Ericsson’s RAN software decoupled from the physical infrastructure, executing entirely on an Intel Xeon 6 system-on-chip.
That silicon integrates advanced vector and matrix instruction extensions tailored for machine learning workloads, removing the necessity for separate GPUs or other discrete accelerators. A hardware-agnostic architecture like this helps operators move innovations from lab prototypes into commercial deployments more quickly and cost-effectively.
Rob Soni, Vice President of RAN Technology at AT&T, commented: “AT&T is leading the charge toward an open, intelligent, and scalable network future by advancing Open RAN and Cloud RAN with AI-native capabilities at their core. This demo highlights how AI capabilities, powered by our next-generation Cloud RAN platform, can be deployed seamlessly to drive innovation and deliver superior customer experiences.”
Accelerating AI workloads in Cloud RAN environments
Infrastructure vendors are building flexible software stacks that operate across multiple deployment scenarios, from cloud-native RANs to purpose-built sites.
Mårten Lerner, Head of Networks Strategy and Product Management, Business Area Networks at Ericsson, said: “Together with AT&T and Intel, Ericsson is demonstrating how our domain expertise combined with AI-native RAN software can drive transformative advancements in both Cloud RAN and purpose-built deployments. Our industry-leading AI-native Link Adaptation serves as the first proof point on this journey. With a hardware-agnostic RAN software stack, Ericsson is committed to offering maximum flexibility and enabling all our customers to benefit from future innovations—regardless of their chosen underlying hardware. This milestone underscores Ericsson’s commitment to helping operators advance their networks by deploying AI functionality across the RAN stack.”
Cristina Rodriguez, Vice President and General Manager of Network and Edge at Intel, added: “This successful collaboration with AT&T and Ericsson showcases the power of Intel Xeon 6 SoC to enable and accelerate AI workloads in Cloud RAN environments. Xeon 6 SoC is architected to handle the demanding compute requirements of AI-native network functions, delivering the performance and efficiency operators need to unlock the full potential of intelligent networks. By providing a flexible, standards-based platform, Intel Xeon 6 enables service providers like AT&T to deploy innovative AI capabilities while maintaining the openness and choice that drive industry innovation.”
The trial maps a clear path toward open, disaggregated architectures. Separating software and hardware enables operators to deploy machine learning where it generates the highest return—whether at the edge, in virtualized Cloud RAN instances, or in purpose-built deployments.
By prioritising interoperability and sufficient processing capacity, network operators can prepare for AI-driven adjustments to transmission quality and resource allocation, improving overall performance across different RAN environments.
Related reading: Qualcomm and others are also preparing telecom infrastructure for AI-native future generations, exploring how processors, software and open interfaces can work together to support increasingly intelligent networks.
Interested in learning more about AI and big data from industry experts? The AI & Big Data Expo runs in Amsterdam, California and London, co-located with other technology events. It provides panels and sessions that cover AI in telecoms, edge computing, cloud infrastructure and related fields.
This report is produced by TechForge Media. Explore upcoming enterprise technology events and webinars directly with the event organisers for further details and schedules.