AI-powered network slicing in 5G networks: real-world applications, techniques, and business model

The combination of network slicing, independent 5G networks , and artificial intelligence is completely changing how mobile infrastructures are designed, operated, and monetized. We're no longer just talking about faster speeds, but about networks capable of adapting to the context, prioritizing critical services in seconds, and creating virtual "mini-networks" tailored to each customer or use case.

In this scenario, initiatives such as Nokia and Amazon Web Services (AWS) with agentic AI , commercial trials by operators such as Orange, du or Telefónica, and enterprise solutions from manufacturers such as Digi or integrators such as Capgemini, show that AI slicing in 5G has gone from being a laboratory promise to a real pillar of the telco business and the digitization of key sectors.

What is network slicing in 5G and why is it now possible?

Network slicing is an architecture that allows the creation of multiple logical virtual networks on the same 5G physical infrastructure. Each of these "slices" behaves as if it were an independent network, with its own resources, security policies, quality of service (QoS) levels, and service level agreements (SLAs).

This segmentation isn't entirely new: it relies on technologies like Software-Defined Networking (SDN) and Network Functions Virtualization (NFV) , which allow for the virtualization of network functions and their programmable management. The difference is that, with 5G, the core has been designed from the ground up to support end-to-end slicing, encompassing RAN, transport, and core.

In practice, an operator can define specific slices for a service, application, industry vertical, or user group , even spanning multiple network domains and potentially multiple operators. Each slice incorporates all the necessary end-to-end elements, from radio access to the core, to achieve a specific business objective.

3GPP has standardized three types of service slices (SSTs) that summarize the major 5G use cases: eMBB for enhanced mobile broadband, urLLC for ultra-reliable, low-latency communications, and mMTC for massive IoT-type communications . From there, each operator can define finer variations tailored to specific customers or sectors.

A key aspect of slicing is the isolation between segments : micro effects (internal changes within one slice) do not impact others, and at a macro level, physical resources can be moved from one slice to another based on demand, thanks to NFV and dynamic orchestration. This allows for guaranteed service concurrency, preventing, for example, a spike in entertainment traffic from bringing down a mission-critical service.

The contribution of agentic AI: from static slicing to the self-adjusting network

Until now, network slicing was seen as a major advantage of 5G , but its commercial adoption has been slower than expected. The main reason is complexity: manually configuring, monitoring, and adapting multiple slices is impractical at scale, and traditional solutions relied on rigid templates and little automation.

This is where agentic AI applied to slicing comes in : specialized AI agents that not only automate tasks but also make context-based decisions, interact with systems, and access real-time data. Instead of static slicing, it shifts to adaptive network segmentation that reacts to the environment almost on the fly.

The joint proposal from Nokia and AWS is a good example. Nokia contributes its advanced network management platform and its AirScale and MantaRay SMO solutions, while AWS provides the AI ​​layer with Amazon Bedrock and Amazon EKS . The AI ​​agents consume data from both the network and open internet sources (traffic, events, incidents, weather, maps, location, etc.) to anticipate demand and automatically adjust RAN and core policies.

These agents operate in several modes— chatbot, on-demand, scheduled, and autonomous —and coordinate via APIs exposed to Amazon Bedrock. This allows operators to deploy context-aware, on-demand premium connectivity services: a slice of network is temporarily reinforced in a specific area during a concert, a sporting event, or an emergency, and then releases resources.

The value of agentic AI, highlighted by operators like Orange and du in their initial trials, lies in its ability to transform network slicing into a genuine revenue driver , not just a technical improvement. Anticipating peaks, orchestrating resources, and closing differentiated SLAs becomes significantly easier when the network "understands" what's happening around it.

AI-powered network slicing use cases in 5G networks

The combination of 5G SA, end-to-end slicing, and AI opens up a tremendous range of enterprise, industrial, and public administration use cases . Some have already been tested in real networks, and others are in advanced proof-of-concept (PoC) testing.

Goal-oriented business and industrial slicing

One of the most powerful scenarios is segmentation focused on business KPIs for enterprises and industrial campuses. Here, the Nokia and AWS solution measures parameters such as bit rate, latency, and availability in real time, and autonomously adapts RAN and core policies to meet agreed-upon SLAs.

Thanks to agentic AI, these slices can prioritize critical applications in manufacturing, Industry 4.0, IoT applications , drones, smart cities, hospitals, energy, transportation, or ports . If an industrial control flow requires ultra-low latency, the system redistributes resources, readjusts queues and QoS policies, and ensures that this traffic takes priority over less sensitive traffic.

In this approach, the network stops operating "on a best effort" basis and starts operating as a catalog of guaranteed services , where each slice is designed with clear objectives (for example, zero packet loss in a robot control system) and AI acts as a continuous supervisor that reconfigures when it detects deviations.

On-demand slicing for emergencies and traffic peaks

Another key use is the dynamic activation of dedicated slices based on external events. By ingesting data on traffic, incidents, social media, weather, and municipal planning, AI can trigger the creation or reinforcement of slices at specific 5G base stations.

During an emergency, for example, an AI agent prioritizes a slice reserved for public safety and emergency services , ensuring critical voice, video, and data even under extreme congestion. At the same time, it preserves quality of service for premium 5G+ and FWA customers using gaming, streaming, XR, or AI applications , adjusting resource allocation without requiring manual intervention.

At large-scale events like concerts or sporting events, AI analyzes the historical network behavior at the venue, infers patterns, and establishes event-specific slicing policies . This ensures additional capacity for VIP areas, payment and ticketing applications, fan interaction, video streaming, and venue operational communications.

Network slicing for SMEs and service environments

The impact isn't limited to large corporations: SMEs in retail, logistics, and services also benefit. A customized slice can ensure robust connectivity for POS systems, mobile apps, real-time inventory, and immersive in-store customer experiences.

In a shopping center or during high-traffic campaigns, the network can automatically readjust bandwidth and priority for certain flows, allowing live video applications, interactive catalogs, or augmented reality without saturating the network of the other tenants.

Furthermore, by combining slicing with business intelligence solutions (for example, dashboards in Power BI tools), companies can cross-reference network metrics with sales and operations data, and understand how connectivity quality influences user experience and billing.

Practical case studies in hotels: guest experience and isolated operations

An interesting real-world example is the one deployed by Telefónica at the Meliá Serrano Hotel in Madrid , where 5G coverage was set up indoors (meeting rooms and bedrooms) and several virtual mobile subnetworks were created using network slicing.

In this scenario, guests connect to a 5G network separate from the hotel's internal systems . Each slice has its own resources, so the traffic generated by guests (streaming, video calls, social media) does not affect critical hotel services (home automation, energy management, security systems, POS, etc.).

The deployment began on a 5G Non-Standalone (NSA) architecture and evolved to 5G Standalone (SA) , demonstrating that it is possible to divide the network into several segments that dynamically allocate resources based on users and load. The case study focused on showing how a hotel can offer differentiated connectivity without having to duplicate physical infrastructure.

Examples in transport, cities and logistics

In transportation, slicing makes it possible to separate autonomous vehicle traffic , traffic management systems, and passenger services into distinct slices. One high-reliability slice is used for low-latency V2X communications, another is dedicated to signaling and monitoring, and a third to onboard Wi-Fi or passenger entertainment.

In smart cities, slices can be deployed for public safety, traffic lights, video surveillance, energy management, or smart parking . Emergency services enjoy a slice with top priority, while other less critical applications coexist in separate segments without interfering with each other.

In advanced logistics, ports and airports, slicing is combined with edge computing to service AGVs, drones, container control systems and warehouse sensors, all orchestrated centrally to meet strict real-time requirements.

Technical architecture: from 5G Core to AI-powered edge

For all of this to be viable, a 5G architecture prepared for slicing from the ground up is needed. The 5G Core Standalone introduces specific functions such as NSSF (Network Slice Selection Function), responsible for selecting the appropriate slice for each user or service, and CSMF, which translates customer needs into technical network requirements.

The user plane relies on User Plane Functions (UPFs) that can scale horizontally thanks to NFV and be deployed at edge locations to reduce latency. Each slice can have dedicated UPF instances, isolated at the resource and security levels, with their own routes and policies.

In the RAN, solutions like Nokia AirScale and MantaRay SMO enable the application of slicing policies at the carrier, scheduler, and radio resource levels. Integration with agentic AI modules and platforms like Amazon Bedrock enables automated decision-making based on radio KPIs, external context, and service objectives.

AI workloads run on Amazon EKS with hybrid nodes , allowing operators to distribute agents and applications across their own infrastructure and the public cloud, while maintaining a unified Kubernetes management plane in both the cloud and edge. Everything is coordinated using standard APIs, facilitating the evolution toward more open and programmable models.

Collaboration between manufacturers (Nokia, Ericsson, Digi), operators (Orange, du, Telefónica, T-Mobile, Verizon, AT&T, Vodafone), and technology partners (AWS, integrators like Capgemini) is crucial for a truly multi-vendor and disaggregated ecosystem . In fact, several recent proofs of concept (PoCs) have focused precisely on demonstrating slicing on disaggregated 5G networks, from radio to core.

Business solutions: routers, multi-slicing, and remote management

Beyond the core and RAN, the success of slicing also depends on what happens at the network edge. Manufacturers like Digi International have incorporated 5G multi-slicing support into their DAL OS firmware, allowing enterprise routers to route traffic across multiple cellular slices simultaneously.

With this capability, an organization can separate and prioritize different types of traffic (video surveillance, industrial control, POS systems, IoT, etc.) without resorting to multiple SIM cards or complex SD-WAN configurations based on VPNs. Everything is managed from the router itself, which automatically detects the slicing configurations offered by the operator.

The solution supports up to 8 slices per SIM (depending on the operator) , allows you to associate specific interfaces with each slice, choose predefined types (low latency, massive IoT, etc.) or custom configurations, and fully integrates with Digi Remote Manager. From this console, you can view the active connections of each slice and monitor deployments in real time.

This approach simplifies the architecture in demanding vertical environments such as smart factories, warehouses, oil and gas, transportation, or smart cities , and at the same time ensures that hardware investments are ready several years in advance, even if the operator's commercial slicing is still in its early stages.

Measurement, SLA verification and perceived quality

One of the barriers to commercializing slicing with strong SLAs was measuring, from the end user's point of view, the performance of a specific slice , since traditional speed test tools only look at the generic Internet connection.

To address this, Ookla and Ericsson have developed a technology that identifies and evaluates the performance of specific 5G network segments. Integrated into a modified version of the Speedtest app, this solution allows for real-time comparison of a standard 5G connection with a dedicated slice of a commercial network.

Operators and enterprise customers can thus independently verify that agreed SLAs are being met by viewing key performance indicators directly from the terminal. This brings transparency to the market and strengthens trust in business models based on guaranteed connectivity.

This functionality is planned for demonstrations at events such as the Mobile World Congress in Barcelona , ​​where it will be shown how KPIs change when traffic is routed through a slice with optimized parameters versus a generic best-effort slice.

Technical, operational and security challenges

Despite the potential, it's not all immediate benefits. Properly implementing enterprise-grade network slicing requires combining expertise in networking, automation, data analytics, device management, and advanced security.

Operators have had to invest billions in 5G spectrum and infrastructure , and are still in the process of recouping those investments. Integrating slicing, agentic AI, and end-to-end orchestration implies new CAPEX and OPEX, as well as profound changes in organization and operating models (moving from manual processes to highly automated AIOps environments).

Security is also a critical issue. While isolating slices reduces the risk of an incident affecting the rest, it's essential to deploy zero-trust architectures, segment-level encryption, continuous AI-powered monitoring , and clear resource-sharing policies. Each slice may require different levels of legal and technical isolation depending on the type of data it carries.

On the other hand, from a regulatory and competition perspective, bodies (such as the EETT in Greece or other national regulators) must establish clear frameworks for pricing, guaranteed QoS, privacy, and data use . Slicing introduces new questions about net neutrality, traffic prioritization, and billing models that are still being defined.

Business perspective: from the network as a cost to the network as a product

The great appeal of AI-powered slicing is that it transforms the network into a programmable service platform . Instead of simply selling generic access, the operator can offer catalogs of predefined or customized slices, with specific SLAs and differentiated pricing.

Among the models being explored, the following stand out: "as-a-service" services per slice with a closed SLA, dynamic pricing based on congestion or service criticality, and rental of slices to virtual operators (MVNOs) or large companies that want to act almost as micro-operators within their ecosystem.

Many stakeholders anticipate that slicing will be one of the major sources of new revenue for 5G networks , allowing for better monetization of the investment made and offering very specific value propositions for sectors such as manufacturing, health, energy, transport, immersive entertainment or public administrations.

In practice, a hybrid model is becoming increasingly common, in which some organizations combine private 5G networks for ultra-critical applications with dedicated slices of public networks for mobile personnel, logistics, branch offices, or coverage in remote areas. This flexibility is one of their major competitive advantages over traditional architectures.

The convergence of 5G SA, network slicing, and agentic AI is driving mobile networks toward a scenario of dynamic, programmable, and business-oriented segments , where each application can have its own "mini-network" with tailored conditions. From industrial automation to massive concerts, and including hotels, smart cities, and SMEs, trials by Nokia, AWS, Orange, du, Telefónica, Ericsson, Digi, and other players show that the potential is already being realized; the challenge now is to scale these models, standardize APIs, and ensure that security, SLA measurement, and regulation keep pace with innovation.