Two years ago, most telecom executives were still weighing whether to put a generative AI chatbot on their customer support site. In 2026, a different conversation is playing out inside the same operators.
NVIDIA’s fourth annual State of AI in Telecommunications survey, released in February 2026 and drawn from 1,038 industry professionals, found that 90% of telecom operators say AI is now driving revenue and cutting costs, and 89% plan to increase AI spending in 2026, up sharply from 65% a year earlier. In a parallel cross-industry NVIDIA survey of more than 3,200 enterprises, telecom had the highest agentic AI adoption of any sector at 48%. The shift is quiet because most of these deployments do not have consumer-facing branding on them. It is significant because it changes what a telco actually does with its workforce.
What agentic AI actually means
Agentic AI is a class of systems in which an AI model is given a goal, access to tools, and permission to act. Instead of a single-turn response, an agent iterates: it plans, calls APIs, checks results, and adjusts. A conversational AI chatbot might explain to a customer why their bill is high. An agent looks up the bill, identifies the specific plan feature causing the overage, checks eligible retention offers against the customer’s tenure, and applies the credit, all in one interaction.
That is the operational difference. It is also the reason telcos, more than most other industries, have leaned in. Communications service providers already run millions of transactions per day through BSS and OSS platforms with well-defined APIs. Those APIs are exactly what an agent needs to actually do something rather than just describe it.
Where telcos are deploying agents today
The current generation of AI-powered telecom solutions typically pairs a foundation model with a set of tools that expose network, customer, and billing systems. Four categories of work have emerged as the near-term focus.
The first is network operations. Agents monitor alarms, correlate them with change tickets, and either resolve routine faults directly or hand off to a named engineer with a diagnosis attached. This is not new science. What is new is that the same agent can act on the findings, not just present them.
The second is customer lifecycle. Agents handle plan changes, upgrades, roaming activation, and simple billing disputes without a human in the loop, then escalate cleanly when the query crosses a defined threshold.
The third is retention. Churn prediction models have existed in telecom for a decade. Agents make them operational. When a model flags a subscriber as likely to leave, an agent can immediately compose and dispatch a personalized offer, then schedule follow-up if the offer does not land.
The fourth is field operations. Agents dispatch technicians, reschedule appointments against real-time availability, and reconcile completed work against inventory systems.
The network operations use case in detail
Network operations is the workload where agentic AI has the clearest early return, and the NVIDIA telco survey quantifies the shift: half of respondents cited autonomous networks as their best-performing AI use case for return on investment, ahead of customer service at 41% and internal process optimization at 33%. That is the first year in the survey’s history that network operations has overtaken customer experience as the top AI ROI category.
The economics behind that finding are straightforward. A single tier-one operator can generate hundreds of thousands of alarms per day, most of which are noise or duplicates. The classic playbook has been correlation engines, rule libraries, and expensive senior engineers whose main job is triage. An agent given access to the alarm feed, topology data, recent change records, and remote diagnostic tools can close the loop on a substantial share of tier-one incidents, and the subset expands as operators codify more of their runbooks in a form the agent can use.
The important caveat is that autonomous action inside a live network is not a place to move fast. Every operator that has deployed agents in production has done so behind approval gates, with tightly scoped tool permissions and detailed audit trails.
The customer lifecycle use case in detail
On the customer side, the interesting question is not whether an agent can answer a billing query. It is whether an agent can complete a plan change without an operator having to expose that ability through a chatbot the customer chooses to use. Increasingly, the answer is that the agent sits inside the retail app, the retention outbound queue, and the care center, taking action across all three surfaces from the same underlying set of tools.
That architecture has second-order effects, and average handle time in care is the most visible one. Early production deployments report double-digit percentage reductions, driven less by the agent talking to the customer and more by the agent doing the back-office work that the human agent used to have to click through.
What is holding scale back
The bottleneck for most operators is not model quality. Foundation models are capable enough for the tasks in question. The bottleneck is the same one that has slowed every wave of telco automation: tool access, data quality, and the operator’s own change control processes. An agent is only as capable as the APIs it can call and the data it can trust.
Governance is the other constraint, and the risk of getting it wrong is now well documented. Gartner has forecast that more than 40% of agentic AI projects across all industries will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. When an agent has permission to spend retention budget, adjust customer records, or trigger a truck roll, the definition of a bad outcome becomes tangible. Most operators are still working out the accountability model for agent-initiated actions and the audit trail regulators will expect to see.
The bottom line
Agentic AI is not a new customer-facing product for the telecom industry. It is a change in how work gets done inside operations that were already partially automated. The operators moving fastest are treating agents as a new class of employee with narrow permissions and detailed oversight, rather than as a chatbot upgrade. That framing is boring, and it is why it is starting to work.
Sources
NVIDIA “State of AI in Telecommunications: 2026 Trends” survey report. Based on 1,038 respondents surveyed September to November 2025. Released February 19, 2026.
NVIDIA blog: “How AI Is Driving Revenue, Cutting Costs and Boosting Productivity for Every Industry in 2026”. March 2026.
Gartner press release: “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027”. June 25, 2025.
