AI-RAN战略的机遇与挑战评估

Author:Richard Webb, Senior AnalystEditor:Ian Kemp, Managing Editor Contents 3The big picture6Key report findings7Section 1:Integrating AI into the RAN10Section 2:AI-RAN drivers and use cases15Section 3:The value of AI-RAN to Open RAN20Section 4:AI-RAN challenges25Additional resources We hope you enjoythe report and, mostimportantly, find ways touse the ideas, concepts andrecommendations detailedwithin. You can send yourfeedback to the editorialteam at TM Forum viaeditor@tmforum.org The bigpicture AI-RAN refers to the full integration of AI into radio access network (RAN)hardware and software in a mobile network. This has the potential to providecommunications service providers (CSPs) with transformative efficiencygains in network resource utilization, spectral efficiency and performance,in addition to enabling new AI services and monetization opportunities. Driven by these potential benefits, CSPs are starting toposition themselves to take advantage of the AI-RANopportunity. In the US, for example, T-Mobile last yearlaunched an AI-RAN innovation center in partnershipwith Nvidia. guaranteed – performance. This gives predictabilityfor time-sensitive applications and data flows, andcontainerization of both workloads, which increasesthe agility of work applications in cloud environmentsregardless of the underlying infrastructure. AI-RANis fully software defined using AI-native principles toenable acceleration of AI and RAN workloads. “AI-RAN at T-Mobile will be all about unlocking themassive capacity and performance that customersincreasingly demand from mobile networks,” said MikeSievert, CEO of T-Mobile, announcing the partnership.“AI-RAN has tremendous potential to completelytransform the future of mobile networks, but it will bedifficult to get right.” AI-RAN is a manifestation of two symbiotic technologytrends: Leveraging AI to enhancenetwork performance Among the challenges, as we see in section 4, arehigh upfront costs and as-yet unproven return oninvestment, high energy consumption, and securityand privacy concerns. Ensuring network performancesupports AI services From an infrastructure perspective, AI-RAN usesa homogeneous, accelerated computing platformwithout RAN hardware components. This meansit can run both mobile network and AI workloadsconcurrently, with deterministic – in other words microservice, meaning the computing layer can sitwith the connectivity layer. Connectivity providersmay not always take control of the AI application,but can be an infrastructure provider for it.This differs from video, for example, in whichcomputing and application sit together. AI-RAN can provide CSPs with a computingfoundation in which the RAN becomes moreautonomous and self-optimizing / self-healing, andintelligent enough to orchestrate radio resourcesbased on real-time patterns, context and intent.With software-defined, accelerated platforms, CSPscan power RAN and AI from the same infrastructure,meaning they can support AI-based services todaywhile futureproofing for 6G. AI-RAN hastremendouspotential to completelytransform the futureof mobile networks,but it will be difficultto get right.”Mike Sievert, CEO,T-Mobile “In the AI world, the compute can sit anywhere,therefore giving an opportunity for connectivityplayers – RAN systems – to be the connectedcomputing infrastructure provider,”says Velayutham. Soma Velayutham, GM for AI, 5G and Telecoms, Nvidia,explains that AI compute is serverless and runs as a Key findings Mobile networks are notjust connective but areincreasingly cognitive – a shiftfrom network as infrastructureto network as intelligence.With AI capabilities becomingmore mature and accessible,adoption of AI in the RAN willincreasingly be a competitivedifferentiator for CSPs as wellas solutions providers. AI-RAN can help CSPsimprove RAN performance, byincreasing connectivity speedand reliability, reducing latencyand energy consumption, andhelping deliver enhanced userexperiences. It can function asa single infrastructure torun both RAN and AI workloads,and provide a platform fordelivering new edge-based AIservices to B2B customers. AI-RAN can enable CSPs toevolve mobile networks fromautomation to autonomousdecision-making, from staticradio planning to dynamicintent-based orchestration.But this requires a mindsetbeyond just embeddingAI accelerators. AI-RANrepresents a transformationin strategy and investment,not just technology. The increasing complexity ofmobile networks means theRAN has become harder todeploy and manage.At the same time, RANprocessing resources arefrequently under-used.Integrating AI into the RANcan give more operationalflexibility by creating oneunified infrastructure. With AI-RAN, CSPs canprovide critical infrastructureand data-driven insights tosectors such as healthcare,finance, manufacturing andretail – fostering collaborativeecosystems. As B2B2Xpartners, CSPs can not onlydrive digital transformationbut also position themselvesas enablers of cross-indus