Testing, Measurement, and Assurance in the AI Era

Sponsored by CONTENTS RCR Takeaways3 Foreword 4 The AI Supercycle The AI upheaval in the network 6 Autonomy, with a hint of caution8 Most developed AI use cases in telco networks16 Conclusion Acknowledgements RCR Takeaways AI is no longer a bolt-on feature; it is changing the network DNA with entirely new traffic dynamics and rising traffic volumes. Operators are working toward Level 4 autonomy with generative AI as the entry point and agentic AI as the next step — but the While telcos focused on network autonomy are tapping it for operational efficiency and cost savings, a growing number ofthem are recognizing its potential for delivering differentiated services and securing new revenue streams. The way operators assure and validate the network fabric in the AI era will separate winners from losers —and determinewhether revenue models, like service level agreement- (SLA) backed services, can deliver on their commercial promise. Legacy assurance frameworks are by design unfit for AI-enriched and AI-native networks. They are characterized by fragmenteddata, siloed tools, domain-specific processes, and reactive workflows with limited value The AI-ready assurance framework must be continuous, cross-domain — and most-importantly, predictive — combining fullvisibility with parallel active and passive testing to anticipate issues before they arise. emergingapproaches network vendorsand operators are developing in response.It draws from first-hand conversations with customer expectations, as new AI-drivenuse cases proliferate across the network. Foroperators to successfully adopt,scale,and support new AI applications,theymust embrace AI-driven testing,monitoring, and assurance that are battle- This report explores this topic from threeangles:the scale of AI’s impact on thenetwork,the growing inadequacy of THE AI SUPERCYCLENotlongago,theword“infrastructure”meantroads,bridges,buildings,power lines,throughmulti-year CapEx commitments,especially in cloud, data centers, GPU as aService, and AI RAN. Over50%telco respondents havemorethan 50 full time equivalents(FTE) dedicated to AI, with two-thirds andwater systems—the physicalfoundationof modern economies.In2026,it rings synonymous with datacenters, fiber routes, servers, and GPUs A joint report by The National Center forEnergyAnalytics and Hamm Institutefor American Energy brings it all home.According to the report, AI infrastructure In2025,50%C-level executives intelcos reported capturing impact fromAI/GenAI, double from 25% in the year TheAI boom is reshaping economies.HarvardeconomistJasonFurmanestimated,and Wired reported,thatdatacenter investments and software The AI-RAN market is expected to growfrom$150 million in 2024 to$816million by 2032 (DataM Intelligence) Within the network, telcos have ramped upAI investment on machine learning (ML),generative AI (GenAI), agentic AI, digital As AT&T CEO John Stankey told The Times,“Ibelieve anybody running a businesstodaywho walks into this momentthinking they’re not going to be disruptedor impacted by AI is being a little naive. Youhave to embrace it. You have to understand •AI investment in telecom is expectedto grow from $17.3 billion in 2024 to$65 billion by 2029, as providers adopt Forthis year,Gartner predicts datacenter spending to soar 31.7%, reaching$650billion,from$500 billion in 2025.WallStreet analysts expect the largest •90%operators report AI is helpingdrive revenue and cut back costs; 60%say their network automation is driven Trend: Opportunity: Implication: AIinfrastructureinvestmentisskyrocketing,withhyperscalersandtelcospouring billions into multi-yearinfrastructure expansion to support next- AsAI budgets expand and measurablebusinessimpacts emerge,operators arepositionedto transition from pilot use Operators that fail to invest now risk fallingbehindon network automation,servicedelivery,and revenue opportunities,as THE AI UPHEAVALIN THE NETWORK hundredsofinteractions,triggeringeast-westdata exchanges,as serverscommunicatebetween themselves.Thisdemandssub-second-levelthroughput and disappear for no apparent reason. Fornetworkoperators,the issues manifestassuboptimalresourceutilization,compromisedtraining,andslow, Nokia, put a number(s) to it at MWC 2026:There are 1.3 trillion annual AI sessions,77 exabytes of monthly traffic, and over100 trillion daily tokens in the network. Massively bursty:AI traffic is massivelybursty,drivenbyuser-initiatedinteractionssuch as chatbot promptsand asynchronous agentic AI automation Lookingback,telconetworkshavemovedthrough a rapid succession ofdominantworkloads—voice,data,media—eachpushing performance, While these issues are rarely catastrophic,overtime they erode the quality ofexperience(QoE),ultimately leading tocustomerchurn.What is worse,theseissuesdo not register as conventionalfailuresor outages that trigger alarms. Properfunctioning of these workloadshingesontheunderlyingfabric.Unfortunately,most networks are not H