How AI is reshaping RAN architectures, deployment models, and the path toward 6G
The panel discussion on AI‑RAN highlighted a mobile industry at an inflection point. While AI has already begun to enhance radio access network performance, the panel agreed that the broader transformation lies in rethinking where intelligence resides, how it is orchestrated, and what future architectures must support. Across operators, vendors, and ecosystem contributors, there was a shared view that AI‑RAN will become foundational to both 5G evolution and 6G design, but only if challenges around trust, distribution, cost, and interoperability are addressed.
AI‑RAN today: early deployments and expanding complexity
Several panellists noted that AI‑RAN currently manifests in multiple forms: AI for RAN, AI on RAN, and AI and RAN, each representing different layers of intelligence. The most mature implementations today focus on centralised AI for RAN optimisation, including energy savings, capacity planning, anomaly detection, and emerging fast root‑cause analysis. These use cases are gaining traction because network complexity has increased dramatically, with operators managing numerous LTE and NR bands, dense cell sites, and diverse traffic patterns.
The panel agreed that AI‑driven anomaly detection and predictive analytics are becoming essential to extracting maximum value from existing assets. Some operators are also experimenting with AI‑based customer experience prediction models, using survey‑driven training data to anticipate sentiment and guide network actions. While these approaches remain early, they illustrate a shift toward experience‑centric network optimisation.
Trust, transparency, and the human‑in‑the‑loop challenge
Despite technical progress, several panellists stated that operator trust in AI‑driven automation remains limited, particularly in Europe. Concerns centre on transparency, explainability, and the risk of AI making autonomous parameter changes without human oversight. The panel emphasised that operators want clear visibility into why AI recommends specific actions and what alternatives were considered.
To address this, the panel highlighted the importance of human‑in‑the‑loop governance, explainable AI, and the use of digital twins to test AI‑driven decisions in a safe virtual environment before deployment. These mechanisms are seen as essential to building confidence and enabling more advanced automation over time.
Edge AI and distributed compute: opportunity meets constraint
The discussion then shifted to AI inferencing at the edge, a topic the panel described as both promising and challenging. As AI‑enabled services such as smart glasses, robotics, and real‑time sensing emerge, low‑latency compute becomes critical. Several panellists noted that operators now have an opportunity to orchestrate both connectivity and compute by placing AI workloads closer to the RAN.
However, the panel agreed that distributed RAN architectures create significant constraints. Many operators have DUs deployed at thousands of cell sites with limited space and power, making large‑scale edge compute difficult. While future silicon platforms may offer more efficient accelerators, the panel stressed that operators must balance opportunity with TCO, sustainability, and the realities of existing infrastructure.
Preparing for 6G: AI‑native design and architectural flexibility
Looking ahead, the discussion highlighted that 6G will be AI‑native, with AI embedded into Layer 1 and Layer 2 algorithms, receiver design, beam management, and sensing‑communication integration. Several panellists highlighted that 6G standardisation is already incorporating AI‑centric assumptions, including the need for flexible architectures capable of supporting rapidly evolving AI models.
A recurring theme was the need for architectural adaptability. With AI technologies evolving faster than telecom standards, the panel stressed that networks must be designed to accommodate new AI workloads without requiring disruptive rebuilds. Smooth migration from 5G to 6G will depend on leveraging existing infrastructure while selectively adding new compute and orchestration capabilities.
Open RAN as an enabler of AI innovation
The panel concluded that Open RAN will play a critical enabling role in AI‑RAN. Open interfaces, modular components, and the RAN Intelligent Controller (RIC) create opportunities for third‑party innovation, faster deployment, and more dynamic optimisation. Several panellists noted that non‑real‑time RIC platforms will make it easier to deploy AI‑driven use cases and push configuration changes across the network.
Open RAN was also described as a catalyst for faster time‑to‑market, particularly for private networks, where AI‑driven orchestration could enable near‑instant deployment of CU/DU components without on‑site engineering.
A sector moving toward intelligent, distributed, AI‑native RAN
The panel closed with a consensus that AI‑RAN is transitioning from experimentation to strategic necessity. Operators see clear benefits in optimisation and automation, but future value will depend on distributed compute, trusted AI governance, flexible architectures, and open ecosystems. As 6G approaches, AI‑RAN will become a defining pillar of network evolution, reshaping how radio networks are built, operated, and monetised.
Panel Details:
(Moderator) James Kirby - Senior Analyst, Analysys Mason
Bernard Bureau - VP, Wireless Strategy & Services, Telus
Simone Tumbarello - Solution Orchestration & Architecture Manager, Adeptic Reply
Ari Kynäslahti - SVP, Strategy & Technology, CTO, Mobile Networks, Nokia, and Board Director, AI-RAN Alliance
Dr. Yue Wang - Chief Technologist, China Telecom
Udayan Mukherjee - Senior Fellow and Chief Architect, Intel