How to reduce RAN energy usage by a quarter with predictive AI

Commercial context

Radio access networks (RANs) can account for 70% of CSPs' total power consumption. Energy usage is rising sharply with 5G densification and traffic growth - increasing costs and endangering sustainability. Yet CSPs’ existing energy-saving methods tend to be based on static settings, vendor-dependent, and lack intelligent decision-making. The result is significant inefficiencies, high operational expenditure, and unnecessary carbon emissions.

The solution

The 'Active Sustainable 5G: Smart RAN energy management' Catalyst addresses the increasingly pressing need for real-time, predictive energy optimization. The project explores how autonomous solutions can continuously learn, predict and address both Open RAN and traditional RAN’s energy-requirements in real time to reduce the network power consumption while securing end user performance is not impacted. The team is developing an AI-based energy management system that forecasts and adjusts RAN power levels based on real-time traffic and network performance data, energy consumption data, environmental factors and network topologies.

During low-traffic periods, the solution can automatically adjust RAN capacity and RAN components’ energy consumption, entering them into low-power states without affecting the network performance. It's designed as an explainable closed-loop system that adapts to network conditions without manual intervention. To that end, the project is employing machine learning within the service management and orchestration (SMO) rApps to dynamically adjust power usage across radio access elements from multiple vendors. The solution is cloud native and vendor agnostic.

Hassan Mohamed, Head of Technology Architecture and RAN at Dhiraagu, a telecoms operator serving the Maldives, explains. "Our solution integrates predictive AI-driven energy management that draws on historical and real-time RAN data. It does this through standardized O1/O2 interfaces, intelligently forecasting traffic patterns to dynamically reconfigure radio parameters via rApps. A dedicated digital twin validates energy-saving scenarios to ensure no degradation in the quality of service.”

The solution’s architecture is built on TM Forum and O-RAN Alliance standards. It therefore supports seamless scalability, interoperability and vendor neutrality - helping to accelerate sustainable 5G deployments. It harnesses multiple TM Forum APIs, including the Resource Inventory Management, Alarm Management, Performance Management and Service Quality Management.

Applications

The team anticipates the agentic AI solution will enable mobile operators to reduce their RAN energy costs by up to 25%. The Catalyst’s simulations and deployments in legacy networks to date have shown RAN energy savings of 15%. Given the scale of energy usage in the RAN, that would have a major impact on operators’ profitability. There is the potential here to improve operator EBITDA margins by up to 3 percentage points by optimizing energy costs. For a large CSP, that could translate into annual savings of US$ 100 million in energy-related operational expenditure. This represents a major ROI in a system that is likely to cost a few millions of dollars.

The solution could also help achieve a 20% reduction in carbon emissions, enabling CSPs to meet ESG goals and ensure regulatory compliance. By helping CSPs reduce their environmental impact, the solution could also enhance their brand reputation. The Catalyst is looking to achieve these cost and emission reductions without impacting service quality, which the system verifies through dynamic AI reconfiguration and compliance with TM Forum standards.

At the same time, the closed loop nature of the solution will help operators to achieve autonomous network maturity level 4, in which operations are automated in specific conditions. At this level, no ‘human on the loop’ is required—only a ‘human in command,’ who retains oversight while the system acts independently. This level of automation, together with AI-optimized configurations, should further drive down CSPs’ costs and increase profitability.

Wider value

A high-level goal of the project is to support sustainable 5G expansion, minimizing the environmental impact of network growth. It is also promoting the adoption of standardized, replicable solutions using TM Forum assets, while demonstrating the operationalization of AI, including reinforcement learning. The Catalyst's champions are four telecoms operators - Axian, GCI, Dhiraagu, Verizon – with operations in Africa, the Middle East, the Indian Ocean and North America.

The team estimates that a 10% reduction in energy consumption could collectively save CSPs globally US$1.8 billion a year. It could also reduce carbon emissions by almost 85 million tonnes annually. As well as addressing immediate business pain points, a comprehensive, sustainable approach to RAN management will make networks more reliable, eco-friendly and cost-effective, helping mobile operators realize the full potential of 5G.