Telecom operators are scrambling to develop autonomous networks managed by AI. Along the way, telcos are risking increasing their costs rather than cutting them, as evidence mounts that AI productivity gains have yet to be translated into lower AI OpEx across the industry, according to consulting company, Bain & Company.
Agentic systems raise expenses if telcos add AI without removing older processes, tools, and workforce costs.
Bain highlighted a scenario where AI agents and the computing tokens they consumer could account for 20% to 30% of a mobile network operator’s (MNO) total operating cost within three to five years, with traditional expenses constituting the remaining 70% to 80%.
Operators are pivoting more into automated network management. In June, 81% of 80 operators surveyed are targeting “Level 4” autonomous networks, involving AI decision-making within specific network domains, by 2030, with a fifth expecting to hit that target as early as 2027, according to industry body, TM Forum.
Costs of Autonomous Networks Market
At Level 4, network autonomy allows systems to monitor network conditions, identify problems, select corrective actions, and check whether those actions worked – alters how operators manage faults, traffic, and performance across complex infrastructure.
According to Bain & Company, the main pitfall is not the price of running AI models themselves, but that operators are adding AI to existing processes without removing the staff, software licenses, outsourced operations, and legacy infrastructure those processors already needed.
Bain named it a “dual operating model” that, at its core, adds cost instead of eliminating it, inevitably further complicating the telecom operating cost breakdown. Instead of replacing expensive processes, AI could simply create a second operating layer around them.
The bigger AI OpEx risk comes when new costs are added without removing older ones. Traditional teams, software licenses, outsourced services and infrastructure may remain in place while operators also begin paying for models, tokens, orchestration platforms, and monitoring systems.
Therefore, a major obstacle to telco OpEx reduction is created.
“Internal chatbots, auto-generated customer call summaries, code assistants, and productivity copilots create value but rarely reshape a telco’s economics,” said Bain.
The consultancy believes AI network operations should therefore be redesigned from beginning to end. Autonomous workflows can continuously detect events, diagnose network faults, select repairs, and verify the result instead of helping employees complete individual steps.
Inference is only one part of AI OpEx. Agent-based systems can also require tool calls, storage, runtime evaluation, observability software and human oversight, meaning the true cost of an autonomous action can extend well beyond the model itself.
Controlling Telecom-Scaled AI Costs
Operators should measure AI’s value by the cost of resolving a complete telecom operating cost breakdown and outcome – such as diagnosing and fixing a network fault – rather than focusing narrowly on individual model queries’ cost. And since agent-based workflows also demand orchestration systems, tool integrations, monitoring infrastructure and human oversight that add expense beyond inference alone.
Vivo in Brazil provides one example of autonomous networks telecom deployments built around an entire workflow. Its self-healing virtualized standalone 5G core detects anomalies, identifies root causes, and automatically applies corrective actions.
Meanwhile, Telefónica said it reduced mean time to resolution for targeted incidents by 30 minutes.
The financial challenge could grow as the autonomous networks market expand. According to Bain, model prices are falling, but overall spending can still increase.
AT&T services show the scale of AI network operations and broader enterprise AI demand. The operator said in July that it processes an average of 45 billion tokens per day and uses an AI gateway to route requests according to cost, speed, and quality.
Bain added AT&T redesigned its orchestration.
“Rather than reducing AI usage, it matched model capability to task complexity, which by AT&T’s account reduced costs by up to 90% while tripling throughput,” said Bain.
That approach could support Telco OpEx reduction by preventing operators from using their most expensive models for every task. Limits on agent runtime, dedicated compute budgets and removing repeated context were also recommended.
China Mobile has also reported operational gains from autonomous networks, including more than a 30% reduction in backend operations and maintenance manpower and an average 30% fall in mean time to repair for faults and customer complaints.
Ultimately, controlling AI OpEx will depend on whether operators remove the processes AI replaces. If old costs remain while agents, models and supporting infrastructure continue expanding, automation could improve performance without delivering the financial savings Telcos expect.
For operators pursuing Level 4 networks, the next AI OpEx phase will be less about proving that AI can run telecom workflows and more about showing that those systems can operate at a lower total cost.
Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Telecom sections to stay informed and up-to-date with our daily articles.
