AI SURGE AND ITS
IMPLICATIONS
FOR 6G
V
Date: 19 February 2026
Programme: 6G
AI SURGE AND ITS
IMPLICATIONS FOR 6G
by NGMN Alliance
Version:
Document Type: Public
Approved by / Date: NGMN Board, 16 February 2026
Public documents (P): © 2026 Next Generation Mobile Networks Alliance . All rights reserved. No part of this document may be reproduced or transmitted in any
form or by any means without prior written permission from NGMN Alliance . The information contained in this document represents the current view held by NGMN
Alliance . on the issues discussed as of the date of publication. This document is provided “as is” with no warranties whatsoever including any warranty of merchantability, non-
infringement, or fitness for any particular purpose. All liability (including liability for infringement of any property rights) relating to the use of information in this
document is disclaimed. No license, express or implied, to any intellectual property rights are granted herein. This document is distributed for informational purposes only
and is subject to change without notice. Readers should not design products based on this document.
CONTENTS
EXECUTIVE SUMMARY
���������������������������������������������������������������������������������������������������������������������4
01
INTRODUCTION����������������������������������������������������������������������������������������������������������
������������6
02 IMPACTS OF AI TRAFFIC ON NETWORKS��������������������������������������������������������� 7
Traffic
Growth������������������������������������������������������������������������������������������������������������������������������
�� 7
Shift in Network
Requirements���������������������������������������������������������������������������������������������������8
03 NETWORK FOR AI
���������������������������������������������������������������������������������������������������������������� 9
Performance vs. Business Value
�������������������������������������������������������������������������������������������������9
Capabilities Beyond Connectivity
�����������������������������������������������������������������������������������������������9
04 AI FOR NETWORK AND IMPLICATIONS
FOR 6G ARCHITECTURE EVOLUTION
�����������������������������������������������������������������11
Network Management
Layer�����������������������������������������������������������������������������������������������������11
Core Network
������������������������������������������������������������������������������������������������������������������������������12
Radio Access Network
����������������������������������������������������������������������������������������������������������������12
Key Challenges and Considerations
����������������������������������������������������������������������������������������12 Implications for 6G
Network Architecture Evolution �������������������������������������������������������������13
05 CONCLUSION & STANDARDISATION FOCUS AREAS���������������������������15
Conclusion
�����������������������������������������������������������������������������������������������������������������������������������15
Recommended Standardisation Focus Areas �������������������������������������������������������������������������15
06 LIST OF
ABBREVIATIONS����������������������������������������������������������������������������������������������17
07 REFERENCES
�����������������������������������������������������������������������������������������������������������������������������18
08 FIGURES
���������������������������������������������������������������������������������������������������������������������������������������19
ACKNOWLEDGEMENTS
����������������������������������������������������������������������������������������������������������������20
EXECUTIVE SUMMARY
This is a pivotal moment in the telecommunications industry,
propelled by the unprecedented AI surge and the beginning of 6G
standardisation. AI is advancing at a rapid pace and will remain a
dominant force, reshaping society far beyond the 6G era.
This document consolidates NGMN’s perspectives on how AI will
likely impact 6G standardisation, providing guidance for
ongoing 6G studies. This study examines three key dimensions:
(1) impact of AI traffic on networks, (2) network for
AI, and (3) AI for network and implications for 6G
architecture evolution.
Impact of AI Traffic on Networks
The rapid proliferation of AI applications – particularly those
with autonomous, task-driven capabilities - introduces significant
uncertainty into future network demand. While the precise
impact of these AI-driven workloads on traffic patterns is
difficult to predict, several factors could materially alter today’s
assumptions:
• Multi-modal AI applications: Services requiring real-time video
exchange may drive substantial traffic growth and shift traditional
traffic patterns.
• AI-enabled devices and use cases: Consumer applications
(. AR glasses) and enterprise scenarios (.
autonomous vehicles) could require frequent upload of
images and video after local processing, increasing uplink
demand and challenging current downlink-heavy network
designs.
• Geographic and device density: AI-intensive areas and
device clusters may experience sharp, localised surges, creating
increasingly uneven traffic patterns.
Given these uncertainties, network design must prioritise
flexibility. Standards Development Organisations should
explore mechanisms that allow semi-permanent adjustments
in uplink/ downlink ratio without requiring major standard
revisions, as well as solutions to enhance the uplink coverage.
This adaptability will be critical to accommodate evolving AI-
driven requirements across diverse devices, networks and regions.
Network for AI
6G should go beyond providing connectivity services to deliver new AI
enabled services and capabilities (. new data exposure), by
designing networks that are more intelligent, flexible, and
trustworthy.
Key design enablers include:
• Flexible (. token-based) charging models reflecting real
resource use.
• Dynamic and intelligent networking for AI agents
collaboration.
• Support for explicit QoS and computing demand from an AI-
based application, to facilitate meeting the required QoS at
minimum cost and environmental impact.
• Enhanced QoS and adaptive policy control to support
traffic routing achieving seamless connectivity.
• Unified data and model frameworks across devices and domains.
• Secure trust, authentication and authorisation mechanisms for
AI agents’ digital identity.
AI for Network and Implications for 6G
Architecture Evolution
AI is expected to be an important network capability for 6G networks,
enabling more efficient usage of network resources, network
automation, intent- based management and intelligent
orchestration. AI could be applicable to all domains and different
layers of the network, including the operation and maintenance.
NGMN expects that 6G will be AI-ready, and the 5G Service-Based
Architecture (SBA) will be considered as the starting point towards
6G architecture.
Challenges and considerations for adopting AI:
• Adoption of AI capabilities should allow agents and large
language models (LLM) to be deployed in a way that avoids
unnecessary impact on the existing architecture. This should
not restrict the possible integration of AI-related features
embedded within network functions (NFs).
• AI interfaces (., A2A, MCP) will complement existing and
future APIs, ensuring readiness for the traffic volumes and
capabilities required by emerging AI services.
• Multi-vendor interoperability frameworks are needed to
ensure secure, scalable, and open ecosystems.
• Deployment strategies must align with cost and sustainability
goals, and validation of real-world performance gains is
essential.
• Continued support for non-AI alternatives if these alternatives are
necessary to ensure reliability, flexibility and openness.
• Coordinated UE–network operation is needed, ., to efficiently
execute AI models in both two-sided and one-sided models.
Recommended Standardisation Focus Areas
• Standardised architecture, protocols, and interfaces
enabling efficient end-to-end support of AI functionalities,
integrated across all domains (RAN, Core, Transport) and all
network layers, including devices.
• Standards that support explicit network QoS and computing
demand from an AI-based application, to facilitate meeting the
required QoS at minimum cost and environmental impact.
• Standards that allow adaptability to support changing
traffic patterns, accommodating uncertainty in the impact of
evolving AI use cases.
• Evolution of the existing (5G SBA) network architecture
should be justified by value driven AI use cases and service
scenarios, ensuring alignment with societal and business
needs.
• 6G standards that support agent-to-agent and agent-to-
network communications.
• Functional and performance requirements for AI capabilities
across the 6G system.
• Establishment of interoperability and trust frameworks to
enable secure, multi-vendor, and multi-agent deployments and
operations (including models retraining, fine tuning).
• Emphasis on the reuse, adoption, or enhancement of “AI interface”
from telco and non-telco worlds where appropriate and
mainstream. (. (A2A) Agent-to-Agent or (MCP) Model
Context Protocol).
01 INTRODUCTION
The rapid evolution of large-scale AI models is driving a paradigm shift toward an “AI-native” era. The proliferation of large language and
multi-modal models is enabling the emergence of AI agents—autonomous, collaborative, and self-learning entities that may outnumber human
users in upcoming years. This shift toward pervasive, agent-driven ecosystems will fundamentally reshape industries, services, and everyday life.
To support this transformation, networks may need to progressively introduce AI features for intent-driven programmability, autonomous
operation, and dynamic compute distribution across central and edge domains. This evolution aims to deliver differentiated connectivity,
high reliability, energy efficiency, and simplified operation, positioning 6G as the best network for AI and a foundation for AI-based applications,
management, and innovation.
As 6G standardisation enters a critical phase, the growth of AI and AI agents presents both opportunities and challenges for mobile network operators
(MNOs). NGMN has outlined key 6G objectives and architectural design principles emphasising innovation across networks, AI, computing,
sensing, modularity, operational simplicity, sustainability, trustworthiness, cloud nativeness, network-as-a-service, automation, smooth
migration, and a disaggregated multi-vendor approach. These principles aim to guide the evolution of networks that are efficient,
sustainable, cost-effective, and socially beneficial [1][2][3][4][5][6][7].
To address the implications of AI on future network design and ensure alignment with NGMN’s objectives, this document examines three
dimensions from an operator’s perspective and highlights recommended standardisation focus areas to support industry alignment:
• Impact of AI traffic on networks
• Network for AI
• AI for network and implications for network architecture evolution
02 IMPACTS OF AI TRAFFIC ON
NETWORKS
TRAFFIC GROWTH
Today, mobile data consumption is dominated by video
applications, accounting for 70-75% of total traffic [8][9]. A handful
of social media and streaming services contribute more than 50% of this
demand.
Although AI applications have grown exponentially, their current
impact on mobile network traffic remains modest – with primary
interactions being text-based [10]. This could change as AI services
proliferate, but predicting the scale of impact remains highly speculative
due to several factors:
• Optimisation of AI Models
AI models are being optimised using techniques such as
quantisation, pruning and reduced token sizes to enable efficient high-
performance inference directly on device. [11]
• Local Processing
More complex AI models are expected to run natively on
device as chipsets evolve with larger and more capable Neural
Processing Units (NPU), faster on-chip memory and cache,
increased RAM allocation for AI workloads and tighter integration
of hardware with AI frameworks and runtime engines.
• Unclear Adoption Curve
End-user adoption curve: it remains uncertain which new
services provide true additional value for end-users, impacting
services adoption, traffic curves and commercial models.
• Regulatory and Privacy Constraints
Several challenges would need to be resolved, for data-heavy AI
features, such as automatic image or video capture via AR
glasses.
Against this uncertainty, the potential impact of AI on traffic
growth needs to be considered in the following aspects:
• Substitution of Current Demand
Multi-modal AI applications are likely to proliferate capturing more
user attention, with smartphones likely remaining a primary
interface. However, it is expected that most video traffic from
these applications will replace existing user behaviour – such as
watching social media video feeds – rather than creating truly
incremental demand.
• Potential Rise of Wearables
AR glasses and similar interfaces could dramatically increase traffic if
they continuously interact with cloud-based AI applications using
video or images. This traffic would be considered incremental,
rather than substitutional, but adoption hinges on overcoming
privacy and security concerns as discussed above.
• Enterprise and Other Applications
Autonomous drones, connected cars, humanoid robots / cobots
and industrial AI use cases could add significant traffic—
provided technological and regulatory hurdles are cleared.
• Uplink Trends
Current uplink demand is moderate, but future use cases such as AI
agents could reverse this trend [10].
AI agents with advanced perception and reasoning capabilities
may reside on smartphones or wearables, continuously
gathering data and interacting autonomously - potentially
generating far more data than humans, subject to battery
capacity and computational power of the device.
However, this shift is uncertain, as many AI agents could instead
operate in the cloud, performing inference and delivering
recommendations to the user.
Future scenarios differ greatly in both likelihood and scale of impact. Use
cases that drive truly incremental video traffic beyond today’s demand
will exert the greatest pressure on networks. While some scenarios
present significant potential for increased demand, they must be
weighed against their likelihood when setting priorities for
network evolution.
This uncertainty makes flexibility a cornerstone of 6G
standardisation – ensuring the network can adapt seamlessly to
diverse and unpredictable requirements.
SHIFT IN NETWORK
REQUIREMENTS
The rise of AI may introduce fundamental changes in both the form
and direction of traffic:
• Machine-oriented Media
Traditional networks primarily carry human- perceivable
content (text, images, audio, video). In contrast, agent-to-agent
communication may involve exchanging models, feature vectors,
latent representations, and other forms of information optimised
for machines rather than humans.
• Uplink-heavy Behaviour
While today’s traffic is mostly downlink-dominated, many AI-
enabled use cases are assumed to reverse this pattern. For instance,
AR glasses with AI may require continuous uplink
transmission of environmental images, and AI-inferenced
autonomous vehicles may upload real-time video and
sensor data more often, in contrast to traditional connectivity
patterns.
6G networks should support these use cases with sufficient
flexibility to increase uplink traffic as a major design driver for
6G networks. For example, increased uplink (UL) slot
occurrences that maximise the UL transmission opportunities to
manage the increased UL traffic expected with new services.
Some of the proposals that are being discussed in industry and
under review in 3GPP are around the definition of flexible and
dynamic downlink (DL) / UL patterns, for example, Full-
Duplex or Sub-band Full Duplex operation. Enhancing UL
coverage is also a desirable feature.
• Regional and Sectoral Variability
The impact of AI traffic will differ across regions and industries. Urban
centers are likely to experience more AI traffic surges than
rural or remote
areas. Certain industries such as manufacturing, transportation,
healthcare, and smart cities may generate higher volumes of AI
traffic. AI-intensive areas and device clusters may experience
sharp, localised surges, creating uneven traffic patterns.
03 NETWORK FOR AI
AI-driven applications impose new requirements on 6G networks,
encompassing not only improved connectivity performance but also
new capabilities beyond connectivity.
PERFORMANCE VS.
BUSINESS VALUE
For performance improvements related to traditional connectivity, it is
essential to validate the necessity of any network enhancements
from a business value perspective in order to avoid unnecessary
investment and resource waste. While network optimisation
can improve user experience to some extent, not all scenarios
require extreme performance gains, as the existing services offered
may not be directly impacted by these network enhancements.
For example, humans generally have a relatively high tolerance for
latency in audio and video conversational interactions through
Internet
—compared with face-to-face communication, users
can typically accept an additional delay of the order of a few
milliseconds without a significant impact on experience.
6G aims to enhance network performance, especially in relation to
network capacity and latency due to the emergence of new services, such
as AI applications. However, in today’s text-based
conversational generative AI services, the dominant factor affecting
response time is not the network latency but the processing delay
of computationally intensive AI models that require specialised and
high-performance infrastructure. In this case, the bottleneck lies in the
AI services and computing infrastructure rather than the network.
Assuming these bottlenecks will be resolved in the future, some
services will require tighter network performance control. For
instance, in the case of conversational AI services for real-time
immersive experience through XR devices (AR, VR and others) the
network throughput and latency requirements will become more
stringent, so that the network Quality of Service (QoS) will need to be
tightly managed to ensure good user experience.
In general, it is therefore important to identify the impact of
network performance on the user experience of an AI service.
Beyond ensuring adequate connectivity performance, the true value of
6G for AI-based services lies in delivering the required
capabilities to efficiently support these new services.
CAPABILITIES BEYOND
CONNECTIVITY
To support AI and AI agents effectively, 6G should integrate
capabilities such as dynamic networking, advanced QoS,
distributed computing, trust management, and intelligent
orchestration.
• New Charging Models
Charging rules for mobile AI services and applications
should reflect their specific demands on network resources. For
instance, a token- based charging model could be
investigated, where tokens correspond to fine-grained units of
resource consumption, such as bandwidth, latency guarantees,
or edge computing capacity. This approach facilitates flexible,
transparent, and scalable transactions among users, AI agents,
service providers, and network operators, promoting fair
cost allocation while incentivising efficient resource usage.
• Dynamic and Intelligent Networking
Future networks are expected to support dynamic and intelligent
collaboration among physical AI agents by enabling the on-
demand creation of intent-driven private networks. These
networks may be short-lived and mission-specific, supporting
scenarios such as collaborative humanoids/robots, drone swarms,
robotic dog swarms, autonomous vehicle fleets, and industrial
embodied AI agents. Compared with static grouping models,
such ephemeral network groups are expected to support
dynamic joining and leaving of agents, adapt to changing
service requirements and
environmental conditions, and minimise manual provisioning
and operational overhead, while dynamically adjusting
membership, connectivity, and performance characteristics based
on task objectives, agent mobility and proximity, real-time service
requirements, and trust and authorisation policies.
• Enhanced QoS Mechanisms
AI services are expanding beyond text to become multi-
modal and it is expected that the communication of different
types of AI-related content will need different treatment. 5G
and previous generations have supported QoS and network
slicing mechanisms to support traffic differentiation, but there
is no means to have clear end user and network-based policies
that enable the routing of traffic into the most suitable connection
(. into the corresponding slice or QoS flow). 6G should
target a much better use of existing QoS and/or slicing
mechanisms and enable advanced policy control with finer
granularity for priority handling and multi-modal information
handling and synchronisation. To achieve this, improving
collaboration with Over-the- Top (OTT) applications and device
manufacturers is needed to face the future demand in the best
way. Furthermore, in case the network involves itself in AI
service tasks in addition to offering connectivity, methods will
be needed to assure the performance of the AI tasks
undertaken, through task level monitoring, measurement and
prediction.
• Edge Computing
AI and AI agent services depend heavily on computing
speed, particularly for low-latency inference. Edge devices / and
user equipment are limited in computing power, while centralised
cloud processing introduces latency and bottlenecks, which
requires careful assessments depending on geo-location. 6G
should support distributed edge computing to enable real-time
processing, collaborative intelligence among agents, and
efficient resource utilisation close to the data source.
• Unified and Distributed Data Framework
Achieving “Intelligence everywhere” requires both data and
compute resources to be available ubiquitously. This implies
transparent data sharing across different domains, which
requires architecture enhancements supported by new protocols
and/or interfaces. AI agents and applications need to share data,
models, inferences, and intermediate results across heterogeneous
devices. Without a unified framework, data may remain siloed and
inconsistent. 6G should introduce an end-to-end data framework to
enable efficient and flexible data, model, and inference sharing,
management, processing, and storage across UE, RAN, Core
Network functions(NFs) and application functions (AFs).
• Trust and Authentication
AI agents acting on behalf of customers require mutual
authentication with networks. Strong encryption and
integrity checks are essential for sensitive prompts and
personal data. Trust frameworks are necessary for agent-to-
agent communication to identify and block malicious AI
content. Compliance and lawful interception capabilities must be
in place to meet regulatory obligations.
• Dynamic and Intelligent Resource Allocation
Adaptive scheduling is needed to handle bursty AI traffic,
prioritising latency-sensitive prompts and inferences while
efficiently utilising shared resources. Orchestration between edge
and cloud AI models enables dynamic workload distribution,
optimising performance, scalability, and resource efficiency,
while dynamic scaling of network functions can help improve
energy efficiency.
• Resilience and Reliability
Mission-critical AI applications—such as those in healthcare or
autonomous control—require continuous availability and
failover mechanisms to maintain user trust.
• AI Traffic Optimisation and AI Agent Interaction
6G should support fine-grained traffic analytics to distinguish
between model updates, inference requests, and agent
communications, enabling optimised management. The network
should also support an AI agents interaction framework that
facilitates seamless interaction between AI agents, the network and
third-party applications.
10
04 AI FOR NETWORK AND
IMPLICATIONS FOR 6G
ARCHITECTURE EVOLUTION
AI is not only a consumer of network resources but also a core enabler
of network evolution. AI will not just help to improve performance of
new networks but also to enable new services and use cases that were
not possible with previous generations, such as digital twins and
sensing.
However, 6G is not just about AI. Some important lessons
learned from 5G show that network evolution should focus
also on aspects such as network simplification and energy
efficiency. These two aspects may contradict AI requirements to
some extent. AI requires the introduction of new network
entities and interfaces which lead to architectural changes,
adding complexity to network evolution. Additionally, AI
engines t ypical ly require more comput ational
resources, leading to some increase in energy consumption.
Therefore, with regard to 6G deployments it is important to
recognize that AI workloads should be deployed where they are
most efficient—across network domains, layers, and physical sites
from central clouds to the edge and even end devices
– and whenever they add some value in terms of network
performance and user experience, hence looking for a good trade-off
between business value, network complexity, energy consumption and
cost.
In 6G networks, AI is proposed to be deeply integrated into
the various layers and domains of network: RAN, transport, core,
and management and orchestration. Depending on the level of
integration, AI could bring more benefits or could pose more
challenges, hence it needs a careful evaluation of what the
requirements are at each domain.
NETWORK MANAGEMENT
LAYER
Network management is a predominant layer responsible for
overseeing all network assets, and the actions it takes can
significantly improve network performance and operational
efficiency given that it controls the entire network, making it
possible to adapt to service requirements and scenario constraints.
For this reason, in this layer, AI should not merely be considered as an
auxiliary tool, but rather as a foundational capability that enables
autonomy and intelligence shifting from rule-based networks
towards autonomous operation.
With this integration of AI, 6G networks can evolve from passive
response to proactive decision- making, supporting intent-
driven management, automation and intelligent orchestration.
• Intent-driven Management
As networks grow more complex, automation becomes
essential to achieve near-autonomous operations. Human
oversight will remain, but operators will express high-level
intents rather than prescribing specific actions. However, special care
is needed for intents with potential conflicting targets, .
performances and energy savings. With high degrees of automation
and use of AI agents, the attack surface of networks and the associated
risk become greater. Failsafe mechanisms should be enabled in order
to both allow the interruption of multiple agents as well as operating
the network in agent-less mode.
• Automation
AI-enabled automation is used to improve resource planning,
anomaly detection, and self-healing, minimising human
intervention and reducing operational costs.
• Intelligent Orchestration
AI and AI agents enable cross-domain orchestration of network,
computing, and storage resources ensuring efficient utilisation
and adaptive service delivery.
• Energy Optimisation
AI has shown benefits for energy saving in 5G and is expected to
deliver further gains in 6G.
CORE NETWORK
The core network domain can benefit from AI adoption in
several areas:
• Network Exposure
As AI is increasingly adopted in third-party software components
(. adopting AI agents), the way in which they interact with
the network is also changing. This requires that network APIs
must evolve to meet these new requirements from third
parties. It is expected that AI agents will consume tools, so it is
required to evolve the exposure layer to manage these new
requirements efficiently.
• Operation and Optimisation
AI tools such as network digital twin can be useful for different use
cases such as root cause analysis, predictive maintenance, capacity
planning, and to assess the impact of new features’ activation or
changes in the architecture. AI integration will also be useful
for the use of core resources and optimise procedures (.
paging, mobility management, and deep packet inspection
(DPI)). However, many of these capabilities may be more related to
implementation aspects rather than architectural
requirements.
• Architectural Evolution
Core network is also required to evolve. Details on network
architecture are discussed in section .
RADIO ACCESS NETWORK
RAN will also benefit from AI integration as this can help to make
more efficient use of radio resources and improve air interface
management. However, not all RAN layers or functions are
expected to benefit equally from AI support, hence it is
recommended to apply AI selectively, focusing
on areas where it will deliver clear value and the benefits (.
performance improvements relative to computational cost) are
notable.
AI is suitable for RAN in those cases where large data volumes
need to be processed and/or need to be resolved, such as for
instance, at layer L2 (MAC-layer). By contrast, functions that
already operate close to the optimal limits with well-
defined, standardised models, or that involve primarily linear
problems—such as channel coding, HARQ procedures or basic
synchronisation—are not expected to see substantial
performance gains from AI-based algorithms. Nevertheless,
AI-based implementations of such functions are not precluded and
may still be considered to improve flexibility, adaptability or
implementation efficiency.
Due to the nature of RAN, implementing AI models at the base
station requires these models to meet strict latency requirements
and performance so as not to impact RAN functions negatively.
This basically implies that AI inference must be executed locally at the
edge to enable real-time operation.
RAN may also require support from UE (two-sided mode), as well as
from the core network or network management to complement AI
processing. This may involve sharing data to/from these entities
or directly sharing model or inference results. Therefore, it is
crucial to assess whether the current interfaces and protocols
allow for optimal communication for AI-transactions. If not, it
will be important to evolve these interfaces or even propose new
ones to make this communication efficient without impacting
overall network performance negatively.
KEY CHALLENGES AND
CONSIDERATIONS
While the benefits of AI integration are significant, its adoption also
introduces several key challenges that require careful
consideration.
• AI Performance in RAN
Some evaluations reveal that many of today’s AI models in
RAN trained on idealised datasets and narrow operating
conditions may exhibit generalised limitations across different
deployment
conditions, leading to context- dependent benefits,
sometimes relatively modest gains, and potential increases in energy
consumption. These findings highlight the need for
comprehensive gain validation in real network environments,
robust cross-domain data collection, and unified AI lifecycle
management and interoperability frameworks.
• Responsible AI
AI technologies introduce potential risks to individuals and
ecosystems. It is essential to ensure the quality, transparency, and
trustworthiness of AI systems. Networks must comply with
regional AI regulations and adhere to operators’ data and AI
ethics.
• Cost and Sustainability
Deployment strategies must align with cost and sustainability
goals, and validation of real-world performance gains is essential.
More specifically, when evaluating the benefits of introducing AI to
enhance an existing network service, the net CO2 impact of AI in
terms of cost and savings should be evaluated in addition to its net
financial impact.
• Non-AI Support
Continued support for non-AI alternatives is required where
these alternatives are necessary to ensure reliability, flexibility
and openness.
• Support for Explicit AI Service Demand
Network support for AI-based services would benefit from
knowing their actual needs, . in terms of network QoS and
computing demand. It would facilitate the network resource
allocation, thereby allowing meeting the required QoS at
minimum cost and environmental impact.
• Considerations for Interconnection with Legacy
Systems
Some existing legacy systems cannot interpret AI-based
requests (., control or management requests in the context of
autonomous networks) due to hardware constraints, which
makes interworking with new systems more challenging.
Therefore, converged management interfaces should be
designed to control both legacy and new systems, while also
considering the gradual replacement of legacy hardware where
required.
• Pace of Technological Change
AI capabilities are advancing very quickly indeed. There is a risk of
standardising 6G functionalities and protocols for AI that will be
out of date at the time 6G networks are deployed.
IMPLICATIONS FOR 6G
NETWORK ARCHITECTURE
EVOLUTION
6G network architecture should ensure the proper integration of AI
across all domains and network layers. 6G is not expected to be a
clean slate/ or disruptive revolution; however, the architecture
should be flexible enough to incorporate new services. Based
on the analysis in previous sections, the following architectural
requirements should be considered:
• Start from SBA
The 5G Service-Based Architecture (SBA) will be considered as the
starting point for 6G architecture, serving as the foundational
framework for the 6G core.
• Intent-based Interaction and Agentic
Communications
AI and AI agents are expected to be pervasive across the 6G network,
enabling intelligent communication and coordination between
key components, especially between external systems such as
UE and third-party applications. These interactions will facilitate
intent-driven management, intelligent network and service
control, autonomous operations, dynamic resource
orchestration, and simplified UE-network interactions.
• AI Agent Framework in Core
AI communication protocols could be adopted independently
while leveraging the current SBA as shown in the figure below. This
may also provide support to AI agents in selective functions to
address new use cases, when justified.
Fig. 1 An Example of AI Agent Framework in Core
• The Model Context Protocol (MCP) is a potential mechanism for
integrating AI agents with network function resources (. NF
APIs). To enable AI agent communication between the network
and the UE—thereby simplifying UE configuration—the MCP
protocol can also be utilised.
• For use cases and procedures where current APIs are suitable, SBI
architectureshouldbeprioritised.
• For new NFs or use cases not supported today, when clearly
justified, the potential use of AI agent-based (agentic)
communications could be assessed in core network architecture,
ensuring multi-vendor interoperability.
• AI in RAN
The evolution from 5G to 6G shifts AI deployment from isolated
single-node processing to a collaborative multi-node intelligent
cluster, enabling dynamic sharing of intelligence and computing
resources across network elements. AI integration in the RAN should
consider two layers: an upper layer, responsible for controlling
and optimising RAN functions, and a lower layer, responsible for
managing radio resources and the air interface. The requirements
for AI/ML implementation differ significantly between these two
layers in terms of data management, real-time execution, and
computing capabilities. RAN node may require
sharing AI models, inference or RAN data with multiple CN
NFs and/or upper layers for cross- domain AI-implementation.
In that case, RAN-CN interfaces will need to evolve to support AI-
services more efficiently. Hence, it is proposed that the
corresponding control plane interfaces will evolve to consider new
transport protocols, such as Quick UDP Internet Connections
(QUIC), or any other enhancements which are more efficient for
future evolution.
• Standardisation for Multi-Vendor
Deployments
3GPP should conduct a comprehensive study to determine
whether communication protocols for AI agents should be
formally standardised or addressed through the future de facto
industry practices —similar to how microservices, container- based
architectures, and for example Kubernetes orchestration have
been adopted in supplier products during 5G core network
development while NFs and NFs’ functionalities / operations were
standardised by 3GPP. The objective is to avoid constraining
future innovation and the emergence of new use cases by mobile
operators. Regardless of the ultimate approach, ensuring
interoperability across vendors and minimising integration
complexity remain critical considerations that must be
addressed.
MCP
Comms
Service Request (.
Intent)
AI GEnerative Service Domain
A2A Comms MCP
Comms
Agent "X" Agent "Y"
MCP Client MCP Client
Agent "Z"
MCP Client
MCP Server
MCP Server
UE
MCP Comms.
Agents accessing Core NF resources via MPC
(. NF APIs)
O&M
Core NFs MCP
Comms
MCP Server MCP Server MCP Server
. AMF . PCF
SBA - no drastic evolution
05 CONCLUSION
& STANDARDISATION FOCUS
AREAS
CONCLUSION
This document has examined the implications of the AI surge
for 6G system design from an operator’s perspective,
focusing on three key dimensions: AI traffic, network for AI,
AI for network and implications for 6G architecture
evolution.
Despite the exponential growth of AI applications, their current
impact on network traffic remains modest, and the scale and
likelihood of future traffic will remain uncertain. This uncertainty
underscores the need for flexibility as a core principle of 6G
standardisation, ensuring the network can adapt seamlessly to
diverse and unpredictable demands.
AI-driven applications will not only reshape traffic patterns—
introducing more uplink-intensive and machine-oriented
communications—but also demand new network capabilities
beyond traditional connectivity.
These include dynamic resource orchestration, intent-driven
management, trust and authentication frameworks, and flexible compute
integration across edge and cloud domains. Meanwhile, AI will become a
key component of 6G networks, enabling autonomous operation,
intelligent orchestration, and proactive decision-making across all
layers of the system architecture.
To achieve these objectives, 6G architecture should ensure
the proper integration of AI across network domains, with the 5G
SBA serving as the starting point for the core network.
Enhancements such as AI agent frameworks, AI agent
communication mechanisms (., current options such as MCP),
intent-based interfaces, and multi- vendor
standardisation will be instrumental in enabling
seamless AI-driven communication and collaboration
between network functions, UEs, and third-party applications.
RECOMMENDED
STANDARDISATION FOCUS
AREAS
The transition towards embracing more AI technologies in
networks presents both opportunities and challenges for
MNOs. On one hand, it promises greater operational efficiency,
service differentiation, and new business models; on the other, it
requires addressing interoperability, trust, and security concerns
across increasingly open and intelligent ecosystems.
To advance this evolution, standards development organisations,
such as 3GPP are encouraged to consider the following areas:
• Standardised architecture, protocols and inter faces
enabling ef f icient end- to- end support of AI
functionalities, integrated across all domains (RAN,
core, transport) and all network layers, including devices.
• Standards that allow explicit demand of the actual needs of AI
services in terms of . network QoS and computing.
• Standards that allow adaptability to support changing traffic
patterns to accommodate the uncertainty on the impact of evolving
AI use cases.
• Evolution of the existing (5G SBA) network architecture
should be justified by value-driven AI use cases and service
scenarios, ensuring alignment with societal and business
needs.
• It is expected that Agent-to-Agent and Agent- to-Network
communication are enabled during the 6G era.
• Standardisation of framework for agent discovery, identity, policy
and trust, enabling secure and interoperable agent to agent and
agent to network interactions across domains and vendors.
• Functional and performance requirements for AI capabilities
across the 6G system.
• Establishment of interoperability and trust frameworks to
enable secure, multi-vendor, and multi-agent deployments and
operations.
• Emphasising the reuse / adoption or enhancement of “AI interfaces”
from telco and non-telco where appropriate and mainstream. (.
A2A or MCP).
• Requirements and architectural support for using 6G sensing
capabilities as a foundational input for a distributed AI data
platform, on top of which AI agents can operate consuming data
from sensing capabilities.
By pursuing these directions collaboratively, the industry can
ensure that 6G evolves into a flexible, sustainable, and intelligent
network—one that supports continuous innovation,
operational simplicity, and meaningful value creation for society,
end-user and industry alike.
06 LIST OF ABBREVIATIONS
3GPP 3rd Generation Partnership Project
A2A Agent-to-Agent Protocol
AI Artificial Intelligence
AF Application Function
AI Artificial Intelligence
API Application Programming Interface
AR Augmented Reality
CN Core Network
DL Downlink
HARQ Hybrid Automatic Repeat Request
L2 Layer 2
LLM Large Language Model
MAC Medium Access Control
MCP Model Context Protocol
ML Machine Learning
MNO Mobile Network Operator
NF Network Function
NGMN Next Generation Mobile Networks Alliance .
NPU Neural Processing Units
QoS Quality of Service
QUIC Quick UDP Internet Connections
OTT Over The Top
RAN Radio Access Network
SBA Service-Based Architecture
SBI Service-Based Interface
UE User Equipment
UL Uplink
VR Virtual Reality
XR Extended Reality
07 REFERENCES
[1] NGMN, 6G Position Statement, Sep 2023,
[2] NGMN, 6G Drivers and Vision, April 2021,
[3] NGMN, 6G Use Cases and Analysis, February 2022,
[4] NGMN, 6G Requirements and Design Considerations, Feb 2023,
pdf
[5] NGMN, 6G Key Messages – An Operator View, Jun 2025,
An-Operator-View_
[6] NGMN, Network Architecture Evolution towards 6G, Feb 2025,
[7] Recommendation ITU-R M. 2160, Framework and overall objectives of the future development of IMT for 2030 and beyond,
[8] AppLogic Networks 2025 Global Internet Phenomena Report,
Reports/GIPR%
[9] Tridens Mobile Data Statistics 2025: Global Usage Trends and Consumption,
[10] Ericsson Mobility Report: GenAI’s impact on network data traffic, Jun 2025,
traffic-
121471#:~:text=While%20most%20mobile%20network%20traffic%20follows%20a%20-
90%3A10,the%20interactive%20and%20content-generation-heavy%20nature%20of%20these%20 applications.
[11] Google Gemini Nano for pixels
08 FIGURES
Figure 1
An Example of AI Agent Framework in Core ....................................................................................................................14
Quan Zhao (China Mobile), Tao Sun (China Mobile), Jose Antonio
Martin Martinez (Telefónica),
Viraj Abhayawardhana (Liberty Global), Guenter Klas (Vodafone).
Editors:
ACKNOWLEDGEMENTS
The following people and companies participated in the development of this publication.
Project Leads: Quan Zhao (China Mobile), David Lister (Vodafone).
Contributors: Eric Hardouin (Orange), Elena Serna Santiago (Telefónica)
1finity, Andrew Tanamas
Amdocs, Andrew D'Souza (until Dec 2025)
Anritsu, Jonathan Borrill Anritsu,
Marsh D'Souza Apple, Mona
Mustapha Apple, Ralf Rossbach
Apple, Kome Oteri
BT, Kevin Holley
BT, Richard MacKenzie
BT, Aaron Walker
China Mobile, Tong Zhang
China Mobile, Xiaoting Huang
Chunghwa Telecom, Che-Wei Yeh CICT
Mobile, Hui Xu
CICT Mobile, Shaoli Kang
CISCO, Fran O'Brien
Deutsche Telekom, Konstantinos Chalkiotis Deutsche
Telekom, Peter Stevens Ericsson, Janne Peisa
Ericsson, Massimo Condoluci
Ericsson, Gunnar Mildh Ericsson,
Torbjörn Cagenius
Fraunhofer Institut, ZoranUtkovski
Fraunhofer Institut, Hauke Buhr
Huawei, PeiyingZhu
Huawei, Xueli An
Huawei, Yan Chen
Huawei, Fei Li Huawei,
Xu Li Huawei, Hui Lin
Huawei, Shaoyun Wu
Huawei, Fang Li Huawei,
Said Tatesh
Institute of Information Industry, Kay Chung
Institute of Information Industry, Yountai Lee
Institute of Information Industry, Frank Su
InterDigital, Ulises Olvera
InterDigital, Sebastian Robitzsch ITRI,
Mitch Tseng
ITRI, Charlie Chang
LG Electronics, Ki-Dong Lee
MTN, Farhan Khan
Nokia, Omar Elloumi
Nokia, Devaki Chandramouli
Nokia, Atte Länsisalmi NTU,
Hung-Yu Wei Orange, Jean
Schwoerer Orange, Arnaud Braud
Orange, Sandrine Lataste
Orange, Dinh-Thuy Phan Huy
Orange, Steve Tsang Kwong U SK
Telecom,Hyeonsoo Lee
Smart PH, Francisco Mangaliman
Telia, Ulf Nilsson
Telia, Anders Rudolphi TELUS,
Aleksandar Prekajski TNO, Ljupco
Jorguseski TNO, ReljaDjapic
Telefónica, Alexander Chassaigne
Telefónica, Jesús Martín Telefónica,
Manuel Nuñez
University of Dresden, Mohammed Elabsi
Vodafone, Chris Pudney Vodafone,
Rishikesh Chakraborty Vodafone, Jörg
Sternagel
ZTE, Feng Xie
ZTE, Yin Gao
ZTE, Yuzhou Hu
ZTE, Jiajun Chen
NEXT
GENERATION
MOBILE
NETWORKS
ALLIANCE
NGMN - Next Generation Mobile Networks Alliance - is a global,
operator-driven organisation established by leading international
mobile network operators (MNOs). As a global alliance of operators,
vendors, and academia, NGMN provides industry guidance to
enable innovative, sustainable and affordable next-generation
mobile network infrastructure.
Key focus areas include Mastering the Route to Disaggregation,
Green Future Networks, and 6G, while supporting the full
implementation of 5G. NGMN drives global alignment of
technology standards, fosters collaboration with industry
organisations and ensures efficient, project-driven processes to
address the evolving demands of the telecommunications
ecosystem.
VISION
The vision of NGMN is to provide impactful industry guidance to
achieve innovative, sustainable and affordable mobile
telecommunication services to meet the requirements of operators
and address the demands and expectations of end users. Key focus
areas include Mastering the Route to Disaggregation, Green Future
Networks and 6G, while supporting the full implementation of 5G.
MISSION
The mission of NGMN is:
• To evaluate and drive technology evolution towards the three
Strategic Focus Topics:
• Mastering to the Route to Disaggregation:
Leading in the development of open, disaggregated, virtualised and
cloud native solutions
• Green Future Networks:
Developing sustainable and environmentally conscious solutions
• 6G:
Providing guidance and key requirements for design
considerations and network architecture evolution
• To define precise functional and non-functional requirements
for the next generation of mobile networks
• To provide guidance to equipment developers, standardisation
bodies, and collaborative partners, leading to the implementation of
a cost-effective network evolution
• To serve as a platform for information exchange within the
industry, addressing urgent concerns, sharing experiences, and
learning from technological challenges
• To identify and eliminate obstacles hindering the successful
implementation of appealing mobile services.
© 2026 Next Generation Mobile Networks Alliance .
All rights reserved. No part of this document may be reproduced or transmitted in any form or by any means without the prior written permission of NGMN Alliance .
