Nokia and Microsoft are combining telecom data products with enterprise analytics and AI tooling in an attempt to move agentic automation from network demonstrations into day-to-day operations.

The companies are integrating Nokia Data Suite with Microsoft Fabric, creating a shared data layer intended to supply AI agents with network, subscriber, radio-frequency, enterprise, and third-party information without operators having to assemble those datasets separately for every use case.

Nokia says the combined system is available now and is designed for multi-vendor, cross-domain telecom environments, including hybrid-cloud and on-premises deployments. Initial applications include voice-over-new-radio assurance, geographical analysis of radio-network performance, predictive maintenance, fault management, automated root-cause analysis, and closed-loop operational workflows.

The proposition addresses a constraint that has followed telecom AI for years, because operators own enormous quantities of useful data but hold those datasets across radio, core, transport, service, subscriber, and IT systems built by different suppliers. Before automation can make a reliable operational decision, the underlying information has to be accessible, current, understandable, and governed consistently.

Data engineering comes before autonomous networks

Nokia Data Suite packages telecom information into reusable data products with industry-specific semantic models and data-quality controls, while Microsoft Fabric supplies storage, analytics, governance, and AI tooling through its broader enterprise platform. Nokia argues that integrating the two can cut data preparation from weeks to minutes for some operational tasks.

That claim will ultimately depend on the state of an operator’s existing systems, but the architecture reflects a more mature phase of enterprise AI deployment. The bottleneck is rarely access to a language model on its own; it is connecting that model or agent to trustworthy operational data, defining which actions it can take, and ensuring that a decision can be traced when something goes wrong.

Telecom networks make those requirements unusually demanding. A poor recommendation in an office productivity tool is inconvenient, whereas an erroneous automated change to a live radio or core network can affect service quality across thousands of customers. Network automation therefore has to combine speed with controls around permissions, rollback, observability, policy, and human intervention.

Nokia’s initial use cases illustrate that graduated approach. In voice-over-new-radio assurance, agents can detect anomalies, perform root-cause analysis, and recommend remedies using a combined view of the network, service, and subscriber. Geo-experience analysis correlates radio sessions with location and performance information to identify coverage or capacity hotspots, while predictive-maintenance agents use historical and live data to identify faults before they affect service.

Agents are moving towards operational authority

Those workloads sit somewhere between conventional analytics and full autonomy. An operational assistant can first surface context for an engineer, then recommend an action, and eventually execute a tightly bounded workflow once confidence and governance have been established. The same data foundation can support each stage, which helps explain why vendors are concentrating on orchestration and data access rather than presenting agent technology as a standalone application.

The partnership also gives Microsoft Fabric a route deeper into telecommunications operations, where specialist network platforms have historically controlled much of the operational data. For Nokia, integrating with a widely deployed enterprise data environment creates a bridge between network information and corporate datasets that sit outside traditional telecom-management systems.

That bridge could become more important as network automation starts to incorporate commercial and customer context. Capacity decisions, fault prioritisation, field-service workflows, energy optimisation, and service assurance can all depend on information beyond the network itself, although combining those datasets increases governance requirements and broadens the consequences of poor access control.

Multi-vendor support will be another important test, because large operators rarely run networks built entirely by one supplier. Automation that works only inside a vendor’s own equipment domain has limited reach, while Nokia says the combined environment is designed to operate across vendors and domains and to support cloud, hybrid, and on-premises infrastructure.

The commercial argument for greater autonomy is straightforward: telecom operators manage large, complicated networks while facing continuing pressure on operating costs, energy use, reliability, and the speed at which new services can be introduced. Automation can remove repetitive diagnosis and workflow execution, but only if the cost of integrating and governing it does not recreate the complexity it is meant to remove.

Nokia and Microsoft are therefore putting the less visible part of agentic AI at the centre of the product — data preparation, semantics, governance, and operational boundaries. If those foundations work across live mixed-vendor networks, agents can take on a larger share of routine operations; if they do not, telecom AI risks remaining a collection of impressive demonstrations alongside systems engineers still have to run manually.