AI is reshaping how future 6G mobile networks are planned, as operators face uncertainty about future capacity and traffic patterns.
The Next Generation Mobile Networks (NGMN) Alliance has published a study outlining how operator networks must adapt to support AI-driven applications. The report gathers operator perspectives to inform ongoing standardisation work within the 3rd Generation Partnership Project (3GPP).
The study stresses that networks must evolve beyond basic connectivity to meet new service demands. NGMN recommends an evolutionary approach that extends current 5G architectures, preserving long-term investment protection and interoperability across vendors and deployments.
Today, most mobile data consumption is driven by video, which represents roughly 70–75 percent of traffic. AI interactions so far are largely text-based and have a limited effect on mobile network load.
But multi-modal AI services could significantly change traffic dynamics, particularly if consumer devices such as augmented reality glasses become common or if enterprises deploy fleets of autonomous vehicles. Those use cases require continuous uploads of environmental images, sensor feeds and telemetry, reversing the typical downlink-heavy traffic profile.
To handle this uncertainty, operators must design flexible networks. Standardisation should explore mechanisms that allow networks to adjust uplink-to-downlink ratios without needing major protocol revisions. For example, base stations may need more frequent uplink opportunities to maximise transmission capacity when devices are sending large volumes of data.
“As 6G standardisation enters a critical phase, the rapid growth of AI and autonomous agents presents both opportunities and challenges for mobile network operators,” said Laurent Leboucher, Chairman of the NGMN Alliance Board and CTO and EVP Networks at Orange Group.
“With the wide variety of future AI use cases, 6G standards must enable adaptability without forcing disruptive architectural changes. Flexibility will be essential to support evolving AI services across devices, networks and regions.”
To monetise new infrastructure investments, telecom operators are considering fresh charging approaches. One proposal is token-based charging, where tokens represent defined units of bandwidth, compute, or edge capacity. This model aims to allocate costs fairly while incentivising efficient resource use by users and operators.
Enterprise AI applications will demand dynamic networking to support collaboration among physical AI agents. Operators should be prepared to offer short-lived, mission-specific private networks on demand to support scenarios like collaborative industrial robots or autonomous vehicle fleets. These networks must adapt to changing service requirements and environmental conditions while minimising manual provisioning.
Edge computing will be critical for low-latency inference. User devices have limited compute, while centralized clouds introduce latency that can be unacceptable for time-sensitive AI tasks.
Distributed computing enables real-time processing near the data source and improves resource efficiency. To make this effective, a unified data framework is needed so AI applications can share models and intermediate results across heterogeneous devices and domains; without it, data remains siloed and limits collaboration.
How 6G mobile networks must evolve to support AI
The shift to 6G will build on the 5G service-based architecture. Network functions must evolve to interact safely and efficiently with third-party software components and AI agents. The Model Context Protocol (MCP) is one candidate mechanism for integrating agents with network resources and exposing model-related context to network functions.
Multi-vendor interoperability remains a primary operator concern. Standardisation bodies must decide whether agent communication protocols should be formally standardised or allowed to emerge through industry practice. Clear interoperability requirements will reduce integration complexity for operators.
Introducing AI into the radio access network requires selective application. Large data flows at the medium access control layer can benefit from algorithmic processing, while functions that already operate near theoretical limits—such as channel coding or basic synchronization—are unlikely to see major gains. Performing inference at the edge is often necessary to meet strict latency constraints and avoid negatively impacting RAN performance.
“The proliferation of AI use cases, particularly autonomous and task-driven agents, is rapidly reshaping how networks are designed and operated,” said Anita Döhler, CEO of the NGMN Alliance.
When assessing new capabilities, operators should measure both the net carbon dioxide impact and the net financial impact to ensure sustainable, value-driven deployments.
Testing AI models in real-world network environments is essential, because algorithms trained on idealised datasets may perform differently across varied conditions. Networks should retain non-AI alternatives to ensure reliability and openness where needed.
The core network will require API evolution to manage third-party software requests and agent interactions efficiently. Many legacy systems lack the hardware or interfaces to interpret new control requests, so operators will need converged management interfaces that bridge legacy and modern systems during transition and hardware replacement cycles.
Trust frameworks are necessary for agent-to-agent communication to detect malicious content and prevent unauthorized actions. Compliance and lawful interception capabilities must remain in place to meet regional regulatory obligations. Strong encryption and integrity checks are required to protect sensitive prompts and customer data.
By aligning early on standardisation priorities, operators can help ensure that future 6G mobile networks are adaptable, sustainable and focused on delivering clear value.
See also: Google’s plans for a major subsea cable to support AI infrastructure
Want to learn more about AI and big data from industry leaders? Attend the AI & Big Data Expo events in Amsterdam, California and London, part of the TechEx series and co-located with related technology expos. The events gather practitioners, vendors and researchers to share insights on AI, data and infrastructure.
Telecoms coverage is provided by TechForge Media. Explore other enterprise technology events and webinars hosted by TechForge Media.