Executive Summary: Agentic AI-Powered 5G-Advanced Slicing
Nokia and Amazon Web Services (AWS) are bringing agentic AI to 5G-Advanced network slicing, moving closed‑loop, intent-based services from PowerPoint to live pilots with du and Orange.
What Changed: Solution Overview
The partners unveiled an agentic AI-powered slicing solution that fuses Nokia’s RAN-to-core slicing, AirScale radio, and MantaRay SMO with AWS’s Bedrock AI platform and EKS Hybrid Nodes to turn external context—events, traffic, maps, weather—and live network KPIs into real-time policy decisions. The result is adaptive, premium slices provisioned when and where they’re needed, without manual reconfiguration.
Why It Matters: Monetization and SLA Assurance
Operators need new revenue from 5G-Advanced while taming unpredictable demand. Static slice policies struggle with concert spikes, emergency incidents, or seasonal surges. Agent-driven, intent-based orchestration can right-size resources dynamically, preserve SLAs, and expose differentiated performance tiers enterprises will pay for. Early field exploration by du and Orange signals that commercial, on-demand slicing is entering a more pragmatic phase.
Solution Architecture and Components
The design blends AI agents, network analytics, and policy control into a closed loop that spans RAN, transport, and core.
AI Agents and Context Fusion with AWS Bedrock
Agentic AI modules ingest open internet data—locations, timetables, incidents, traffic flows, maps, and weather—alongside operator telemetry. Using AWS Bedrock to access foundation models and coordinate reasoning, agents infer demand hotspots and intent, then propose or execute slice policy updates. Modes include chatbot-assist, on-demand triggers, scheduled routines, and fully autonomous operation, all via API integration.
End-to-End Control, Policy, and Assurance
Nokia’s advanced slicing coordinates across AirScale base stations, transport, and 5G core with MantaRay SMO acting as the control point for RAN policies, QoS profiles, and service-level objectives. The loop measures live KPIs such as throughput, latency, and congestion, and adapts policies to keep SLAs in bounds. This aligns with 3GPP’s analytics- and policy-driven operations in 5G-Advanced and can complement functions like NWDAF and NEF for observability and exposure.
Cloud-to-Edge Deployment with EKS Hybrid Nodes
AWS EKS Hybrid Nodes allow carriers to run AI agents and network workloads on-premises or at the edge while maintaining unified Kubernetes management. Bedrock supplies managed model access and toolchains to build specialized agents, with inference orchestrated against operator and contextual data. Nokia’s Edge Slicing extends these capabilities, placing applications closer to users on high-capacity, low-latency paths.
Early Adopters and Launch
du and Orange are the first operators exploring the capability on live networks, with demonstrations at Mobile World Congress 2026.
du: Critical Communications and Premium Experiences
du’s interest centers on responsiveness for critical communications and premium consumer experiences. In emergencies, agents can prioritize first responders around incident zones by boosting capacity and resilience at affected cells. For commercial tiers, on-demand optimization can preserve experience for high-value subscribers when contention rises.
Orange: Intent-Driven Slices Across Use Cases
Orange is testing how intent-driven policies can anticipate demand and tailor slices across diverse use cases, from industrial control to immersive media. The goal is to shift from static profiles to proactive, context-aware service delivery at venue, campus, and city scale.
Why Now: 3GPP Release 18 and Traffic Variability
5G-Advanced (3GPP Release 18) formalizes AI-native operations just as operators face stalling consumer ARPU and rising traffic variability.
From Capability to Productized Slices
Network slicing has existed on paper for years, but monetization faltered due to manual provisioning, long lead times, and opaque service tiers. Agentic AI reduces time-to-action, enabling slices that come online for a stadium gate opening, a storm moving across a metro, or a port’s shift change—then wind down as conditions normalize.
Operational Leverage via Closed-Loop Automation
Closed-loop automation can cut waste from overprovisioning while improving SLA attainment. With intent expressed in business terms—venue load, fleet dispatch, priority users—the system translates objectives into policy, enforces them, and verifies outcomes with continuous telemetry.
Priority Use Cases for Commercialization
Initial commercialization will favor scenarios with clear intent, measurable KPIs, and immediate business impact.
Enterprise and Industrial SLAs
Across campuses and city zones, agents can track latency and throughput and adjust RAN scheduling, QoS, and prioritization to meet contracted targets for manufacturing control, smart logistics, energy grids, hospitals, and ports.
Public Safety and Emergencies
Triggered by incident data, the system can uplift capacity and prioritize first responders at selected sites, preserving reliability and preemption when congestion peaks.
Mass Events and Managed Venues
For concerts and sports, agents forecast load from event metadata and historical patterns, then pre-stage capacity and apply differentiated tiers for broadcasters, payments, venue ops, and VIP experiences.
Premium Consumer and FWA Tiers
On-demand optimization can protect gaming, streaming, XR, and AI assistant sessions for premium mobile and fixed wireless access customers during surges or adverse weather.
Strategic Implications for Operators
Agentic AI turns slicing into a programmable revenue engine, but only with the right guardrails, integrations, and metrics.
Monetization and API Exposure
Define productized slice tiers with clear KPIs, credits, and remedies; integrate with BSS for dynamic charging. Consider exposure via GSMA Open Gateway/CAMARA-style APIs so enterprises and developers can request performance on demand, bounded by policy.
Data Governance and Risk Management
External context boosts accuracy but requires rigorous sourcing, privacy controls, and model governance. Establish provenance checks, PII handling, regional data residency, and safety policies for autonomous actions.
Operations Model and Human-in-the-Loop
Adopt human-in-the-loop for high-impact changes at launch, graduating to autonomy as confidence grows. Align fault, performance, and security workflows so AI recommendations translate into sustainable runbooks and SRE-like practices for RAN.
Standards Alignment and Ecosystem
Map agent functions to 3GPP analytics and policy interfaces and ensure SMO alignment. Validate multi-vendor interoperability across RAN, transport, and core to avoid lock-in while preserving closed-loop performance.
Next Steps and KPIs
A phased roadmap reduces risk while building a repeatable, monetizable slicing practice.
Near-Term Actions and Pilots
Select two to three high-ROI venues or corridors; codify intent taxonomies; instrument KPI baselines; and pilot agentic policies with bounded scopes. Integrate Bedrock-based agents with SMO and assurance systems, and wire charging for event-driven boosts. Build a governance board for model evaluation and action approval.
KPIs to Watch
Track SLA attainment, time-to-provision, cost per delivered Gbps at peak, premium ARPU uplift, churn reduction in premium tiers, incident resolution time, and automation confidence scores. These determine whether agentic slicing is ready to scale beyond showcase deployments.
