Google Cloud unveils AI framework for autonomous telecom networks

Google Cloud has outlined a new artificial intelligence framework designed to help telecommunications operators analyse, predict and respond to network problems, with the longer-term goal of creating highly autonomous networks.

The framework combines a live digital representation of a telecoms network with machine-learning models and AI agents. Google says the approach could allow operators to move beyond traditional monitoring and automated rules towards networks capable of identifying problems, determining their likely causes and recommending or carrying out corrective action.

At the centre of the system is a network digital twin. Rather than being a static map of network infrastructure, Google describes it as a temporal graph that continuously represents the physical and logical state of the network, including how that state has changed over time.

This historical information can allow AI systems to examine both current network conditions and previous states when investigating an incident. Google says this can help with root-cause analysis and give network operators greater insight into how faults develop.

Predicting network failures

A major component of the framework is the use of Graph Neural Networks (GNNs), a type of machine learning designed to work with data where relationships between individual elements are important.

For telecoms networks, those relationships can include connections between routers, interfaces, services and traffic flows. Google says GNNs can be used for anomaly detection, root-cause analysis and predictive maintenance.

The technology could, for example, identify patterns associated with equipment failures or connection problems before they result in a service disruption. It can also be used to analyse potential handover failures involving fast-moving equipment and allow operators to consider actions such as load balancing or rerouting traffic.

Another proposed application is what-if analysis. Operators could simulate scenarios including fibre cuts, sudden traffic increases or configuration changes and examine how the resulting disruption could propagate through the wider network.

Moving towards autonomous operations

Google Cloud says the framework is intended to support the industry’s move towards Level 5 autonomous network operations under the TM Forum model.

The concept involves progressively reducing the amount of manual intervention required to operate complex networks. Google’s framework is designed to provide AI agents with access to network information while retaining a digital environment in which proposed changes can be evaluated before they are applied to production infrastructure.

Google has also demonstrated how trained models can be deployed for real-time inference through its Gemini Enterprise Agent Platform. In a root-cause analysis scenario, an anomaly can be submitted to an inference endpoint, which returns network entities identified as likely causes of the problem.

The company has been developing the approach for some time. At Mobile World Congress 2026, Google described its move from AI systems that primarily provide insights towards agents capable of sensing network conditions, reasoning about problems and taking action.

Wider industry push

Google’s work comes as telecoms equipment manufacturers and cloud providers increasingly explore AI-driven network automation.

In June, Nokia and Google Cloud announced an expanded partnership to integrate Gemini-powered AI agents into Nokia Assurance Center. The companies initially developed six specialised agents covering areas including event triage, anomaly analysis, network performance and remediation.

The developments highlight a broader shift within the telecoms industry towards AI-assisted and eventually autonomous network operations. Rather than relying solely on predefined rules and manual troubleshooting, operators are increasingly looking at systems that can analyse large volumes of network data and determine the relationships between individual faults and their wider impact.

Google’s latest framework brings together the data, machine-learning and AI-agent components required for that approach, while the digital twin provides a way to test potential changes before they affect live customers.

For telecom operators, the objective is ultimately to move from reacting to network failures after they occur towards predicting problems and, where appropriate, automatically taking action to prevent or minimise service disruption.