Google Cloud and Nokia Ship an AI Agent That Runs Telecom Networks
Google Cloud and Nokia on Wednesday introduced a jointly built artificial-intelligence system for telecommunications operators, the latest attempt to bring autonomous network management out of pilot programs and into commercial deployment.
The system, built on Google’s Gemini models, sits inside operators’ network operations centers and performs work that engineers once did by hand: detecting congestion, rerouting traffic, tuning radio parameters and predicting failures before they happen. Nokia supplies the network management platform and the telecom domain expertise; Google supplies the AI models and the cloud infrastructure. The two companies said the system has moved from concept to paying customers, with early deployments underway at operators in Europe and Asia.
Telecom has been one of the most discussed and least realized markets for AI. Operators spent years testing self-organizing networks, intent-based management and automated optimization tools, and every major equipment vendor launched an AI initiative. What has been missing, executives say, is a system that operators trust enough to put in control of live traffic, where a mistake means dropped calls or a city losing connectivity.
Google’s pitch is that its models have crossed that trust threshold. The Gemini-based agents are trained on network telemetry and designed to explain their decisions in plain language, so a network engineer can see why the system moved capacity from one cell tower to another. The system starts in advisory mode—recommending actions that humans approve—and operators can gradually grant it more autonomy as it proves itself.
The commercial push reflects a broader pattern at Google: applying its AI stack to industries where data is abundant and software has lagged. The company has made similar moves in healthcare, manufacturing and retail, pairing Gemini with industry partners who provide domain knowledge. Telecom, with its billions of connected devices and chronic shortage of specialized engineers, is a natural target.
Nokia brings the operator relationships. The Finnish company’s network management software runs in a large share of the world’s mobile networks, and its executives have said the AI partnership is central to a strategy that has shifted from hardware toward software and services. Nokia, which absorbed Alcatel-Lucent a decade ago and has cut thousands of jobs since, sees automation as a way to sell more software to the same customers.
The economics line up with the industry’s pain. Operators face stagnant revenue and rising data demand, and they have been trimming engineering staff even as networks grow more complex with 5G, private networks and thousands of small cells. Routine tuning and fault management consume a large share of network teams’ time, and automation that absorbs that work has an immediate cost case.
Competitors are converging on the same opportunity. Microsoft struck a multiyear AI partnership with AT&T last year, and Ericsson, Nokia’s chief rival, has its own AI-driven management software. The difference Google and Nokia claim is the depth of the models: Gemini’s ability to reason over telemetry streams and explain its actions, rather than apply fixed rules.
The two companies declined to name the operators in the early deployments, citing confidentiality, but said the first commercial contracts were signed this quarter. Google has positioned Gemini as the model of choice for enterprises embedding AI in operations, and the Nokia partnership is one of its highest-profile wins in a regulated industry that has historically been slow to adopt cloud services.
The market is large enough to matter. Operators globally spend tens of billions of dollars a year running their networks, and the software that manages them is a multibillion-dollar category that has grown only modestly in recent years. Automation is the lever that could change that: if AI agents take over routine management, operators may spend more on software precisely because they are spending less on people, a trade that vendors are counting on.
Analysts said the real test is deployment. Telecom operators are conservative buyers, and network management is safety-critical; a system that performs brilliantly in a lab and fails once in production can set adoption back years. Google and Nokia are betting that explainability—the ability to show why a decision was made—will overcome that conservatism, and that operators will accept partial autonomy before full autonomy.
For Nokia, the deal is part of a broader transformation away from its hardware roots. The company has said it wants software and services to account for a growing share of revenue, and AI network management is the clearest example of that ambition. For Google, it is a foothold in one of the last major industries where its cloud and AI offerings have not yet become standard.
Neither company is alone in the field, but the Google-Nokia system is among the first to reach commercial deployment with a major equipment vendor, and its progress will be watched as a measure of whether telecom AI was overhyped or simply slow.
