As artificial intelligence and radio access networks (RAN) increasingly converge, ZTE believes operators should distinguish between two different approaches: “AI for RAN,” which improves the radio network itself, and “RAN for AI,” which uses RAN computing resources to run general-purpose AI applications.
Sun Yue, deputy general manager of RAN marketing products at ZTE, said the two approaches should be evaluated separately from a computing-power perspective.
“AI for RAN” uses AI to make the radio network more efficient and powerful, with benefits potentially extending across individual sites. By contrast, “RAN for AI” uses RAN computing power for general-purpose AI applications, but Sun said its business model is not yet mature and demand is currently limited to a few niche scenarios.
ZTE therefore believes operators should pursue a decoupled, heterogeneous architecture for the two approaches.
“Our belief is clear: keep the two apart with a decoupled, heterogeneous architecture — the best path to a good return on investment,” Sun said.
For operators, ZTE sees the near-term value primarily coming from AI for RAN, through lower operating expenses and higher revenue. Sun said ZTE has made progress across four areas based on a dual-layer intelligence architecture operating at both site and network levels.
AI-powered Massive MIMO can increase cell capacity by 15 to 20%, according to ZTE, while AI-based energy saving can reduce power use by 10 to 15%. In operations and maintenance, mean time to repair (MTTR) has fallen by 20%.
The company also says differentiated experiences can help operators move beyond selling data volumes. “By offering tiered, differentiated experiences, operators can move from ‘selling gigabytes’ to ‘monetizing experience,’ lifting ARPU by 5 to 20%,” Sun said.
ZTE also expects AI services to place greater requirements on bandwidth and latency. Sun said the company will work with operators to provide differentiated, deterministic experiences for AI applications.
From selling gigabytes to monetizing experience
Sun said operators face a “critical paradox” in which data usage is increasing while average revenue per user remains flat. ZTE’s response is to shift from data-centric to experience-centric operations, using differentiated and deterministic connectivity to support specific services and scenarios.
For video streaming, this could mean instant 1080p HD playback. For live broadcasting, it could mean reliable connections without interruptions, while competitive gaming requires ultra-low latency.
“Differentiated and deterministic experience requires tailoring network capabilities to every real-life user scenario,” Sun said.
ZTE’s approach involves three steps: defining key services such as HD video, interactive live streaming and cloud gaming; observing high-demand scenarios such as crowded stadiums, marketplaces and commuter routes; and dynamically allocating network resources according to service requirements and wireless conditions.
The company has applied this approach in several operator collaborations. With China Mobile, ZTE worked to secure live-streaming experience at a prominent crystal marketplace serving global customers. According to Sun, the initiative benefited approximately 10,000 streamers within six months and improved streaming quality for both fixed booth livestreamers and roaming livestreamers.
ZTE has also worked with Thailand and Indonesia operators on experience-guarantee explorations involving major events, universities and outdoor festivals. Sun said in guaranteed scenarios, users were 20% more likely to experience speeds above 5Mbps than in non-guaranteed scenarios, without impacting the experience of background users.
Looking ahead, Sun expects AI applications, including digital and physical Agents, to create greater demand for deterministic RAN.
“Looking ahead, AI applications like digital and physical Agents will drive a massive demand for deterministic RAN, making it the core competitive battleground for 5G-A and 6G,” he said.
ZTE sees Agent Teams reshaping network operations
ZTE’s vision for AI-driven network operations also extends beyond an AI copilot. Sun said the company sees an “Agent Team” as a way to coordinate specialized agents covering areas such as network optimization, fault management, energy efficiency and operations.
The approach combines real-time data, domain knowledge, AI reasoning, operational feedback, and execution tools. According to Sun, this can support proactive detection, root-cause analysis, solution generation, human-authorized execution, and evaluation of outcomes.
Each completed workflow can also feed back into the system, enriching its knowledge, skills and strategies.
ZTE sees several high-value applications for this model. These include detecting faults and quality issues proactively, identifying root causes, generating solutions, and executing approved actions. The Agent Team can also support differentiated service assurance by identifying quality degradation, assessing service impact, and applying targeted optimization policies.
For customer complaints, the system can distinguish between common and individual issues, coordinate solutions and return results to the customer-service platform.
Energy efficiency is another application. Sun said the Agent Team can use traffic load, coverage, KPIs, and equipment status to dynamically coordinate energy-saving policies while protecting user experience.
The broader objective, according to ZTE, is to move network operations from individual task automation toward a continuously learning and self-evolving system.
“Ultimately, the value of the agent team collaboration is not simply to automate individual tasks, but to transform network operations into a continuously learning and self-evolving system,” Sun said.
By connecting expert knowledge, real-time network intelligence and cross-domain execution, ZTE believes the approach can help operators move from reactive issue handling toward proactive service assurance, while improving operational efficiency, user experience and sustainable network growth.
For RAN computing power used for third-party inference, Sun said the capability remains on ZTE’s roadmap and that the business model is still being explored.
“Our philosophy is simple: deploy where value is proven today, and explore tomorrow’s opportunities step by step,” he said.
