AI in Networks Market (2026-2036)

The growth of this market is mainly driven by the widening divergence between network traffic growth and operator revenue, rising network energy cost and the quantified savings delivered by artificial intelligence-driven control, expanding operator cost reduction programmes with quantified targets and deadlines, and growing network scale and parameter count beyond manual operating limits. However, declining and flat capital expenditure across major operator groups, limited availability of standards-based control interfaces in deployed networks, and the absence of verified cost cases combined with tightening change control after documented outages restrain the growth of this market.

Furthermore, rising operator commitment to autonomous network maturity targets, expanding deployment of agentic artificial intelligence in production network operations, and growing public funding directed specifically at artificial intelligence-native networks are expected to offer growth opportunities for the stakeholders in this market. However, limited data granularity and historical depth in network operations, the high and unquantified power requirement for inference at cell sites and edge locations, the narrow supply of network artificial intelligence skills against unmatched training standards, and tightening vendor restriction regimes limiting supplier choice remain major challenges impacting the growth of this market. Additionally, rising automated resolution rates across operator network operations, the growing separation of network energy consumption from traffic growth, growing concentration of traffic intensity rather than connection growth, and expanding open radio access network commitments without published cost cases are prominent trends in this market.

The AI in Networks Market comprises the software and services through which artificial intelligence and machine learning are applied to the planning, configuration, optimisation, assurance, energy management and security of communication networks. The market includes network intelligence and analytics platforms, domain intelligence applications for radio access, transport, core and enterprise networks, automation and closed-loop orchestration software, RAN Intelligent Controller applications, service assurance and experience analytics, network security analytics, network digital twin and planning software, and the consulting, integration, model development, managed operations and support services that deliver them. The market is measured on a software and services basis. Network equipment and port value, fronthaul and backhaul capital expenditure, probe, telemetry and test hardware, software-defined networking and virtualisation platform licences, network infrastructure built to carry artificial intelligence workloads, observability of artificial intelligence systems, and operations analytics outside the network domain are excluded. The market ecosystem extends from telecom operators, enterprise network owners and hyperscale providers to radio, transport and enterprise equipment vendors, network software and operations support system suppliers, artificial intelligence platform and model providers, test and telemetry vendors, systems integrators and managed service providers, and the standards bodies, industry alliances, regulators and security agencies that specify and govern this work.

The market is experiencing strong growth as network traffic continues to rise faster than the revenue that funds it, which has moved optimisation software from a discretionary purchase to an operating requirement. The Ministry of Industry and Information Technology recorded mobile internet traffic of 219.7 exabytes across the first half of 2026 in China, up 17.7% year on year, alongside telecom business revenue of CNY 887.3 billion, down 2.1%. Ofcom recorded United Kingdom mobile data of 1,257 petabytes per month in Connected Nations 2025, up 18%, against United Kingdom telecoms revenue of GBP 34.7 billion, down 0.8%. The Telecom Regulatory Authority of India recorded wireless data usage of 84,424 petabytes in the quarter ended 30 June 2026, up 8.30% quarter on quarter, while average revenue per gigabyte fell 5.07% to INR 7.13. The Ericsson Mobility Report places total mobile network traffic at 203 exabytes per month at the end of 2025, rising to 515 exabytes by 2031 at a compound annual growth rate of 17%, against subscription growth of 1%. Capacity additions alone cannot close that gap, and optimisation software has become a condition of serving demand profitably.

The maturing of standardised management and control interfaces is further strengthening market growth, as it allows artificial intelligence functions to be procured as managed objects rather than as vendor-specific features. The 3GPP specification TS 28.105 on artificial intelligence and machine learning management reached version 20.0.0 on 25 September 2026, standardising a model lifecycle of training, testing, optional emulation, deployment and inference, with normative interfaces for federated, distributed, reinforcement and transfer learning, inference emulation before deployment, explainability, and capability configuration explicitly covering next-generation radio access network energy saving, mobility and load balancing, while stating that the model algorithm itself is not in scope for standardisation. TS 28.104 on management data analytics reached version 20.1.0 on the same date, defining standardised analytics capabilities across coverage, service level, fault management, energy saving, mobility, maintenance and resource domains. The O-RAN ALLIANCE completed Specification Release 5 on 8 June 2026, introducing artificial intelligence and machine learning workflow services spanning the Non-Real-Time and Near-Real-Time RAN Intelligent Controllers. Operators can therefore specify artificial intelligence functions in tenders as managed objects with defined interfaces, which opens the market to application suppliers that do not also supply the radio.

Procurement demand is being reshaped by the arrival of independent validation of autonomous operation, which has given buyers a comparable measure of capability. TM Forum launched its Autonomous Network Level Assessment Validation programme in June 2025 and had awarded almost 100 certificates by DTW Ignite 2026, including 40 new certificates to 17 communications service providers at that event and 35 certificates to 21 operator networks at Innovate Asia in November 2025. A TM Forum survey of 80 operators reported at DTW Ignite 2026 found that 20% expect to reach Level 4 or above by 2027, 81% are targeting Level 4 or above by 2030, and 75% planned to increase autonomous network investment by 2026. Rakuten Mobile was validated at Level 4 under the RAN energy efficiency optimisation scenario in February 2026, and Zain Kuwait published scores of 4.0 for that scenario and 3.9 for RAN fault management. Suppliers are now bidding against named scenarios and validated levels.

Another major growth driver is the cost of network energy, which has produced the clearest published returns of any application in this market. Bharti Airtel reported 133,198 megawatt hours of electricity saved in the financial year ended March 2026 through machine learning optimisation placing low-load radios into sleep mode. Vodafone UK reported on 11 March 2025 that artificial intelligence reduced daily power consumption of 5G radio units by up to 33% at trial sites across London. Rakuten Mobile reported an approximately 20% reduction in radio access network energy use from its autonomous energy efficiency solution across tens of thousands of base stations. Deutsche Telekom reported that energy intensity fell approximately 16% in 2025 to 48 kilowatt hours per terabyte, with artificial intelligence-based data centre cooling management delivering efficiency gains of up to 33%. The GSMA reported in Mobile Net Zero on 14 July 2026 that operator emissions fell 13% between 2019 and 2024 while data traffic more than quadrupled. Energy applications consequently carry the shortest approval cycle in this market and are funding the platform investment that later applications depend on.

Demand for autonomous fault handling is increasingly being met through agentic systems operating inside live networks rather than through reporting tools. Deutsche Telekom disclosed on 5 October 2026 gross savings from artificial intelligence and automation of approximately EUR 1.1 billion by 2027 against 2023 outside the United States and approximately EUR 2.5 billion in indirect cost savings by 2030, with a target of more than 90% automated fault resolution in network operations by 2027 and its RAN Guardian Agent cutting response time to network strain events from several hours to about one minute. China Mobile raised the autonomy level of its network operation centre from 3.2 to 4 on the TM Forum assessment, reducing manual operations by the equivalent of 5,500 full-time roles by the end of 2024, cutting backend operations manpower by more than 30% and average fault and complaint mean time to repair by 30%, across deployments covering 300 million subscribers in Guangdong and 330 million in Xinjiang. Verizon reported more than 70 million autonomous network configuration changes in a year. Contract value is shifting accordingly from reporting and visibility tools toward systems that execute changes under governance.

At the same time, the growing use of accelerated computing inside radio access networks is reshaping the network intelligence ecosystem and creating new opportunities for suppliers able to combine inference with radio processing. NVIDIA agreed to invest USD 1 billion in Nokia at a subscription price of USD 6.01 per share, announced on 28 October 2025, covering the Aerial RAN Computer Pro platform, CUDA hosting of Nokia radio software and Spectrum-X Ethernet switching, with T-Mobile US committed to field evaluations from 2026. SoftBank reported uplink user throughput gains of approximately 20% from artificial intelligence uplink channel interpolation, approximately 13% in high-mobility conditions from sounding reference signal timing prediction, and approximately three times the spectral efficiency from 16-layer multi-user multiple-input multiple-output on graphics processing units. Nokia launched a commercial artificial intelligence-native radio access network platform on 15 July 2026 built on its anyRAN software with NVIDIA accelerated computing and Marvell silicon, offering a graphics processing unit capacity plug-in for installed basebands, a standalone accelerated node and a cloud deployment path, with pilots from the end of 2026, commercial availability in 2027 and pricing as a software subscription. The AI-RAN Alliance reached 132 members by 26 February 2026 across three working groups, having published four foundational documents. Radio and inference workloads sharing a platform is creating a supplier category that neither traditional radio vendors nor cloud providers previously occupied.

This reflects the emergence of independent autonomy validation as the procurement benchmark, which favours suppliers able to map products to named scenarios over those positioning on feature breadth.

This indicates that enterprise adoption is being carried by licence bundling rather than discrete purchasing, while the practical constraint has shifted to the quality and consistency of existing network data.

This points to procurement weighting governance, gating and rollback controls above autonomy itself, which lengthens sales cycles for closed-loop functionality and raises the value of auditable control frameworks.