Bringing base-station intelligence into 5G operations must be a priority for CSPs

Commercial context

What makes the 5G era distinct is the vastly increased demand for diverse services, heavy data traffic, and fast-moving user expectations. The commercial opportunities here are self-evident. Yet there is a snag - many CSPs still rely on centralized systems that cannot react quickly enough to local variation in demand or quality. This creates familiar problems: service performance looks the same for every user, operations teams spend time on repetitive tasks, complaint volumes remain high, and diagnosis takes too long. Networks work hard but fail to deliver differentiated value or meaningful efficiency gains.

This strain is becoming critical as AI workloads rise across industries. Existing architectures cannot process these demands at speed because intelligence sits far from where issues emerge. Base stations collect data but do not reason or act on it. Every decision must travel through central systems, creating delay and limiting the CSP's ability to guarantee quality. The gap between network complexity and operational capacity widens with each new service.

These pressures also create financial consequences. CSPs face sustained OPEX from manual troubleshooting and long mean-time-to-resolution. Energy consumption climbs as density grows. Customer retention weakens when high-value activities, like video conferencing, gaming, or live broadcasting, perform inconsistently. The result is a market where it is difficult to differentiate, innovate, or protect revenue.

The Smart 5G with intelligent computing Catalyst responds to these challenges by repositioning where intelligence resides. Instead of relying on distant systems, it places AI directly at the network edge. This approach transforms the model from reactive command-and-control to proactive, adaptive, and self-optimising. It aligns with the goal of building AI-native networks and supports operators as they move toward 5G-Advanced and, eventually, 6G.

The solution

The project introduces a cloud-edge collaborative architecture that brings intelligence closer to the user. Its core innovation is the Intelligent Computing Board, deployed inside base stations. This component enables real-time perception, analysis, and decision-making at the point where service issues occur. It strengthens local processing, reduces backhaul load, and allows fast responses to congestion, anomalies, and mobility events. This creates an immediate performance benefit that centralized systems cannot match.

At the platform level, the team developed the Unified Wireless Communication Intelligent Computing Platform (UWCICP). This system brings together data, models, orchestration, and service exposure within a single operational layer. It builds a closed loop that moves from perception to decision to execution with minimal external intervention. This loop supports both local autonomy and coordinated optimization across wider clusters.

The architecture uses a structured model layer that hosts multiple AI agents. Each agent focuses on a specific operational outcome. The QoS assurance agent detects service conditions, evaluates priority, and allocates resources in real time. This allows operators to guarantee performance for high-value users and services, even in dense or mobile environments. It keeps latency low and stabilises traffic during handovers or contention.

The energy-saving agent analyzes traffic patterns to reduce power use without affecting experience. It identifies redundant configurations, powers down components during low-load periods, and adjusts strategies across sites and scenarios. This produces consistent OPEX savings and a measurable reduction in carbon impact.

The network optimization agent handles diagnosis and performance engineering. It uses machine learning to recognise anomalies, locate issues, and recommend or execute fixes. It removes manual steps that previously required human analysis, reducing repetition and freeing engineers to focus on higher-value work. Its capabilities also support digital-twin-based simulation and predictive maintenance, allowing faults to be addressed before users notice them.

Across these functions, the platform exposes capabilities as standardized services. This supports rapid deployment, multi-vendor integration, and alignment with TM Forum assets such as eTOM, Autonomous Networks requirements, intent frameworks, and TMF921 API. It keeps the architecture open and evolution-ready, enabling CSPs to add new agents and business logic as 5G-Advanced matures.

Application and wider value

This Catalyst provides gains that span performance, efficiency, cost reduction, and customer experience. These gains have been validated in live environments including multiple provinces in China including Shanghai, Shaanxi, Sichuan and Jiangsu, and multiple scenarios such as high-speed rail and major venues. The results show that an AI-native edge architecture can create measurable improvements at scale.

Performance improves immediately when high-value services receive priority during congestion. Tests show guaranteed low latency and zero packet loss for premium users. In high-mobility scenarios, delay for tasks such as WeChat transfers drops by 20%, helping maintain stability during travel or large events. The platform also reduces the proportion of poor-quality cells by 25%, boosting overall service consistency. Complaint resolution improves as well: handling efficiency rises by 30%, and the rate of successful resolution increases by 10%. These outcomes strengthen trust and reduce customer churn.

Efficiency also rises across operational teams. The automation of diagnosis and optimization cuts mean-time-to-resolution from hours to minutes. AI agents complete tasks that previously required human analysis, allowing skilled engineers to focus on strategy rather than triage. In field deployments, staff efficiency improves by 12%, supported by better access to observable data and shared knowledge. The network optimization agent also reduces manual site visits, lowering labor demands and increasing repeatability across regions.

The energy-saving agent delivers significant financial and environmental value. By adjusting power use during low-traffic periods, operators achieve 10–15% annual energy savings across sites. Average reductions reach 785 kWh per cell each year, lowering OPEX and helping operators meet sustainability targets. These savings do not compromise experience, as the agent adjusts behaviour dynamically based on predicted load. This shift makes energy management a continuous, automated process rather than a manual or reactive task.

For CSPs rolling out these capabilities across national networks, the financial upside is substantial. Based on current resources and user bases, the solution can generate more than ¥560 million in equivalent annual revenue through differentiated services and improved performance. It can also reduce OPEX by ¥125 million through automation and energy savings. These figures demonstrate how AI-native architectures unlock long-term value rather than short-term efficiency alone.

The platform also enables a more flexible commercial model. By supporting differentiated assurance, operators can introduce tiered service levels for events, campuses, enterprise sites, and premium users. This shifts connectivity from a homogeneous commodity to a tailored resource aligned with customer needs. It also supports rapid creation of new services through the API-exposed capability layer, giving operators a faster route to market. Together, these gains show how edge intelligence, when embedded into the network architecture, can transform both operations and commercial outcomes.