报告：如何在移动网络中部署人工智能？

Content Page 31Overview.................................................................................................................................. 2AI in the telecom landscape: AI for networks and networks for AI............ 53Strategy for business growth with AI.........................................................................3.1 Wide range of AI-powered solutions 64The path to AI-native RAN..............................................................................................4.1 Evolution journey ahead of 6G4.2 The three stages of AI integration with RAN4.3 Data strategy: High-quality data drives performance4.4 Selecting the best AI technology depending on the use case 85AI-driven hardware evolution.......................................................................................5.1 AI RAN hardware processing approaches in the industry 106Deployment architecture: Maximizing value..........................................................6.1 Balancing performance and efficiency6.2 Combining the advantages of both centralized 137Future strategy and emerging opportunities......................................................... 8Recommendations and Key takeaways.................................................................... 18Authors....................................................................................................................................... 1 Artificial Intelligence (AI) is transforming how networksare built, operated, and monetized. For CommunicationsService Providers (CSPs), AI is no longer optional; it isa strategic imperative to manage growing complexity, This report outlines Ericsson's approach todeploying AI in mobile networks, groundedin real-world experience and technicalleadership. It distinguishes between “AI fornetworks”—enhancing network perfor-mance and automation using AI —and strategy for two hardware architecturetracks (purpose-built and Cloud RAN),Ericsson ensures that AI is executed whereit delivers the most value—whether at theedge for ultra-low latency or centrally fornetwork-wide optimization. These AI-powered solutions improve user experience, Ericsson's AI RAN strategy spans bothcentralized (rApps) and distributed (radiosite) deployments, enabling non-real-timeand real-time automation and new AI usecases. With a deep integration of AI into For CSPs shaping the future of telecom,this report provides a strategic lens on howto scale AI efficiently, maximize return oninvestment, and build future-proof networks AI in the telecomlandscape: AI fornetworks and 2 AI transforms networks in two fundamental ways:first, it enhances network efficiency and second,it enables entirely new AI-powered applications thatdemand consistent, high-performance connectivity, Ericsson distinguishes between these twotransformative roles through clearterminology: “AI for networks” refers toleveraging AI technology to enhance net-work operations and performance, while differentiated connectivity. The “AI fornetworks” approach represents the mostpromising path forward for managingtoday’s exponentially growing complexityof mobile networks—from surging traffic Strategy for businessgrowth with AI 3 while a network’s self-adaptiveapproach reduces operations complexity.AI technologies allow networks to under-stand these intents or CSP's businessobjectives, process large data sets, makereal-time decisions, handle conflictsresolution, and optimize the network AI in networks improves performance andunlocks new growth opportunities. Ericssonenvisions AI as a key technology enabler forbuilding high-performing programmable Intent-driven means that CSPs canspecify their desired network outcomeswithout detailing how to achieve themor mentioning the specific configurationsrequired for implementation. This could 3.1Wide range of AI-powered solutions Ericsson is a pioneer of AI in telecom. InRAN, the AI journey began with 4G andEricsson's AI-native approach, laying thefoundation well before the advent of 6G.AI technologies used in Ericsson solu- Digital twinis used for simulation andtraining of the AI models and to deploy Generative AIis used to support networksoperations, software development and tions for automation1include generative AI,digital twin, neural networks, reinforcementlearning and AI agents. The goal is to deployAI in networks efficiently and scalable withan architecture that maximizes the return Agentic AIwill enable networks to makeautonomous decisions based on the Reinforcement learningoptimizes radioresource allocation to improve spectrum The path toAI-native RAN 4 4.1Evolution journey ahead of 6G The AI RAN journey that began in 4G isnow rapidly evolving toward AI-nativearchitectures in advance of 6G deployment.This transformation represents a funda- examples include mobility optimizationto improve handover speed and reduceddropped calls, AI MIMO sleep mode to 4.2The three stages of AI integration with RAN AI started a