电信人工智能：2025年第二季度市场状况

With AI, it’s more a case of whatisn’thappening The GSMA is a global organisation unifying the mobile ecosystem todiscover, develop and deliver innovation foundational to positive businessenvironments and societal change. Our vision is to unlock the full powerof connectivity so that people, industry, and society thrive. Representingmobile operators and organisations across the mobile ecosystem andadjacent industries, the GSMA delivers for its members across three GSMA Intelligence is the definitive source of global mobile operator data,analysis and forecasts, and publisher of authoritative industry reports andresearch. Our data covers every operator group, network and MVNO inevery country worldwide – from Afghanistan to Zimbabwe. It is the most GSMA Intelligence is relied on by leading operators, vendors, regulators,financial institutions and third-party industry players, to support strategicdecision-making and long-term investment planning. The data is used as We invite you to find out more atwww.gsma.com Our team of analysts and experts produce regular thought-leadingresearch reports across a range of industry topics. www.gsmaintelligence.cominfo@gsmaintelligence.com Authors Kirti SadanaSenior Research Tim HattHead of Research andConsulting Ankit SawhneySenior Research 01 Executivesummary 02 03 Summary Watch out for edge inference 03 You can’t manage what you can’t measure 01 AI’s pervasive nature across industries makes its impact bothobvious and, by the same token, at times hard to pinpointdepending on what it is being used for. The telecoms industry isno different in this respect. If 2024 was about establishing thestrategic rationale for AI, 2025 is about assessing progress. Tothis end, GSMA Intelligence has evolved this research series toestablish a baseline that can be tracked over time for the industry, Running inference at the edge has several selling points foroperators compared to processing AI workloads in the cloud.Cost savings (potentially 30–40%), lower compute latency,network resilience, data sovereignty and energy sustainabilityare all part of the value proposition. Early inference use casesfor operators concentrate on on-premises for a range of Early wins versus slower burns 02 AI is a long-term game. Operator deployments will followa phased implementation, with easy wins achieved beforenewer, potentially more complicated integrations are tried.Customer care is among the ‘low-hanging fruit’, accountingfor around 50% of the AI deployments tracked. Networksfollow, at around 20% (GSMA Intelligence tracks only publiclyannounced deployments, so the true number is likely to behigher). Of the AI deployments analysed, 75–80% (depending AI in numbers 10-20% 47% 60:40 62% Telco AI with arevenue objective Expertise andexperience Live versustrial/planned Customer care Customer care (call centresand other sales touchpoints)accounts for nearly half ofthe telco AI deploymentstracked by GSMA Telco AI deployments haveso far primarily focused oninternal cost savings andefficiencies. The use of AIto drive new business is lessprevalent, remaining at 10–20% of deployments tracked.Many of the revenue models 60% of the AI deploymentstracked have already beenlaunched by operators aspart of their day-to-daybusiness. The remaining 40%are in the trial or planningstages. Unlike with 3G/4G/5Gnetworks, where trials linearly Across many AI investmentrequirements, 63% ofoperators prioritisedupskilling. Network capacity(62%) follows closely behind.Upskilling points to therequirement to constantly 02 developments 03 Recent developments: news flow Commercial Telkomsel and Perplexity establish a strategiccollaboration to expand AI adoption A1 and Cognigy bring AI into corporatecustomer service (read more atA1) Implications of news flow Commercial Technology •Know your purpose.AI deployments continue atpace among mobile operators. Most deploymentsare being driven by cost savings, with the remaining15–20% targeted at revenues through new or •At the edge.Early commercialisation efforts areleveraging edge compute as a way to run AIinference at a lower cost compared to the cloud,which handles the bulk of training workloads. This •Faster regulation?Regulatory initiatives (e.g. theEuropean Commission’s Apply AI Strategy) indicate •Rule makers versus rule takers.AI regulation isa fast-forming but fluid area. Operators can helpshape the regulations rather than simply being •GPU dynamics at play.Nvidia has a significantfirst-mover advantage in GPUs; Intel and AMD areless represented in telco AI activity. The challengefor the latter two is that Nvidia has in-built scale •AI-native RAN?Embedding AI in mobile RAN is agiven, but how deep AI is integrated remains anopen question. All the major equipment vendorshave signalled support for the deeper approach, but •Sovereign AI.This is likely to become an evenstronger current in 2025. It represents a keycompetitive advantage for operators