Telecom Operators Test Whether AI Networks Can Move From Opex Cuts To Revenue

Telecom AI Is Moving Beyond Cost Control

Communications service providers are using generative and agentic AI first to make networks cheaper and more reliable to run.

The operational case is already concrete: automation can handle complex network processes, improve performance indicators and reduce operating expense when operators prepare data, select use cases, integrate systems and validate changes carefully.

The industry benchmark cited for this shift is TM Forum’s Autonomous Networks Project.

Its maturity model has six levels.

Most service providers still place themselves at Level 1 or 2 across the whole network, while some leading operators have validated Level 4 processes in narrower domains.

Those Level 4 examples show why AI operations are becoming more than a lab exercise.

China Mobile has reached Level 4 in network operations centers using agentic and generative AI.

The cited outcomes are a 30% cut to operations-and-maintenance staffing requirements and another 30% improvement in the mean time to repair faults and customer complaints.

Rakuten Mobile has reached Level 4 in RAN energy efficiency, with an expected 20% improvement in radio access network energy efficiency.

Swisscom has also validated Level 4 in IP transport, where the result is cost savings and faster service-extension timelines.

Networks Become Part Of The AI Infrastructure Stack

The next question is whether operators can sell AI-era infrastructure, not just use AI internally.

Service providers control distributed connectivity and compute assets that are relevant to customers looking for data localization, low latency, security and proximity to users, devices and data.

Cisco chair and chief executive Chuck Robbins described the opening as a path for operators to define their role in monetizing AI.

His point was that AI needs bandwidth, distributed infrastructure, secure connectivity and closeness to users and data, all areas where telecom operators already have assets.

That argument also changes how old network sites are viewed.

Robbins pointed to central offices and mini data centers spread across telecom networks as locations that could be reused in the AI era.

Instead of treating those sites only as legacy infrastructure, operators can assess whether they support edge compute, AI inference, sovereign AI services or GPU-as-a-Service.

Enterprise Demand Will Test The Edge Thesis

AT&T’s Andy Forester framed the moment as different from earlier edge-compute cycles.

He said edge compute had been like a use case searching for a market for the last 10 to 15 years.

The difference now is enterprise demand for workload-placement choices: inside company facilities, closer to edge sites, in dedicated AI data centers or through hybrid designs.

That creates a practical test for communications service providers.

Private 5G and multi-access edge computing were often sold with a build-it-first mentality.

AI workloads are less forgiving because architecture depends on the use case, data location, latency profile, security model and business outcome.

The final checkpoint is whether operators can turn autonomous-network progress into customer-facing AI revenue.

The present business case is opex reduction.

The stronger telecom infrastructure story will require paid services around connectivity, edge capacity, security policy and workload placement for enterprises adopting agentic AI.