• AI is expanding value for telcos from data transport to data compute. Specifically, telcos’ distributed networks position them to play in edge compute, critical for latency-sensitive AI use cases.
• But this opportunity contested, with hyperscalers extending their platforms toward the edge and retaining control of developer ecosystems and workload orchestration.
• Nevertheless, ‘edge AI’ represents a high-value opportunity for telcos across consumer and industrial applications where latency, reliability and network awareness are critical.

Limits of the traditional telco model built on selling data traffic is structurally constrained, resulting in flat revenues and margins in the face of continued traffic growth. AI workloads increase demand for both bandwidth and compute, but do not align with legacy pricing models, reinforcing the need for new monetization approaches.

Value is shifting further from network to compute and orchestration in the AI era, providing an opportunity for telcos to redefine their role beyond connectivity. While operators benefit from proximity to users and distributed infrastructure, capturing value depends on translating these assets into services rather than capacity.

Competition and telco monetization hinges on securing control points in the emerging compute layer, where value is increasingly concentrated.

Hyperscalers such as Amazon Web Services, Microsoft Azure and Google Cloud are extending their reach toward the edge, challenging telcos on both infrastructure and developer ecosystems.
At the same time, players like NVIDIA are shaping the AI compute stack, while industry initiatives such as GSMA Open Gateway aim to help operators expose and monetize network capabilities.
Telcos that can integrate connectivity, compute and APIs into cohesive enterprise offerings stand to unlock new revenue pools, while those that fail to move up the stack risk being relegated to low-margin connectivity providers. 

For telcos, the shift to compute is a critical transition that will determine whether they can deliver and capture AI-driven value or continue to see it flow to others.

Limitations of the Pipe Model

The traditional telecom model is structurally constrained. While data traffic has grown exponentially, revenues and margins have remained largely flat with data a commoditized market.

Telcos face three key challenges: 

1) Limited monetization upside as connectivity scales traffic but not profits;

2) Value capture shifting to hyperscalers in the AI era, while telcos continue to provide mostly underlying connectivity; and

3) Weakening customer ownership constraining pricing power, upstream value capture

These limitations in the context of AI workload demands actually highlight a growing opportunity for telcos to expand networks from passive infrastructure to programmable platforms better aligned to support latency sensitive, AI-driven applications.

From Network to Platform in the AI Era

AI is changing the nature of network demand, with infrastructure evolving from simply moving data to supporting how it is processed, prioritised and acted upon in real time. This shifts the network from passive infrastructure toward a more programmable, service-oriented platform.

Defining the AI ‘platform telco’

A platform telco moves beyond connectivity to operate as a service-enabling layer within the digital ecosystem.  This model is defined by three core capabilities:

1) Exposure of network capabilities via APIs, allowing developers and enterprises to directly access functions such as quality of service, identity and location;

2) Enablement of third-party services, including AI-driven applications and enterprise workloads deployed across network and edge environments;

3) Integration of compute, network and data, into a unified, software-defined system capable of supporting real-time workloads 

Early implementations are already visible through initiatives such as GSMA Open Gateway and AWS Wavelength. The latter illustrates hyperscalers extending their developer platforms into telco networks, while the former represents an industry effort by operators to standardise and expose network capabilities such as identity, QoS and location via APIs. However, while Open Gateway establishes the foundation for telco-led platforms, large-scale monetisation and control remain at an early stage. While telcos are beginning to expose network capabilities via APIs, this does not yet translate into full platform control, as most AI workloads and developer workflows remain anchored in hyperscaler environments.

Why this becomes important in the AI age

Historically, networks were optimized for high-volume, best-effort data delivery. In contrast, many emerging AI use cases require real-time response, dynamic resource allocation and continuous interaction with the physical environment, especially with respect to inference. While AI training remains centralized in hyperscale data centers, it is widely anticipated that the majority of AI compute will shift toward inference as models move into large-scale production deployment, with latency and proximity to the end user becoming critical design considerations.

Latency requirements are tightening. Applications such as industrial automation, autonomous systems and real-time video analytics often require sub-10ms while cloud architecture brings higher round-trip delays dependent on distance. At these thresholds, performance degradation can render many applications unusable.

At the same time, network traffic continues to scale without a proportional increase in monetization. Global mobile data traffic has grown at around 40% CAGR over the past decade, while ARPU has seen only low single-digit-percentage annual growth in the past five years. AI workloads risk accelerating this imbalance, as they increase demand for both bandwidth and compute without fitting neatly into traditional pricing models.

As a result, networks must support where and how data is processed, prioritized and acted upon in real time. This is driving the transition from static transport layers to programmable, software-defined systems capable of supporting AI-driven services.

 Key enablers of the platform shift

The transition toward platform telcos is being enabled by three key developments: 

• Agentic AI: Autonomous systems are increasingly interacting with network infrastructure in real time, requiring dynamic orchestration of compute and connectivity resources;
• Edge compute: Processing data closer to the user reduces latency and enables real-time AI inference, aligning application provider needs with telcos’ distributed infrastructure footprint; and
• Network programmability: APIs and software-defined control layers allow network capabilities, such as quality of service, identity and location, to be exposed and consumed as services.

Together, these developments enable networks to evolve from static infrastructure into flexible, service-oriented platforms capable of supporting AI-driven workloads. 

Control, Competition and Monetization

The shift toward platform telcos is about expanding where value is created in the AI network stack by leveraging natural control points. 

Emerging control points
The digital stack is defined by five layers:

• Infrastructure (telcos and vendors): physical networks, spectrum and deployment
• Connectivity (telcos): access, data transport and quality of service
• Cloud (hyperscalers): centralized compute, storage and AI training
• Platform/API layer (contested): hyperscalers currently dominate orchestration and developer access
• Applications (AI providers): user-facing services and monetization

Historically, telcos have controlled the infrastructure and connectivity layers, while the majority of value has been captured at the cloud and application layers. AI does not fundamentally change this dynamic at the top of the stack but introduces a new potential control point in the middle: the edge or network platform layer. 

To understand where this shift creates opportunity, it is important to first examine how value is distributed across the digital stack.

Shifting Value Across the Digital Stack (Traditional vs AI Era)

This layer governs how network capabilities are exposed, how workloads are orchestrated and, for many latency-sensitive and real-time applications, where AI inference is executed. It is also where edge compute, APIs and real-time network intelligence converge.

Telco and hyperscaler competition and collaboration

The strategic question is not where compute sits, but who controls how it is orchestrated across cloud, edge and device layers. Currently most of it lies with hyperscalers, though their relationship with telcos is both competitive and collaborative. Telcos depend on hyperscalers for AI infrastructure, developer ecosystems, cloud-scale compute, while hyperscalers depend on telcos for connectivity, last-mile access, edge presence and local infrastructure.

For the growing universe of real-time, latency sensitive and network-aware inference AI workloads, we see two possible outcomes for network control:

1) Hyperscaler-led where compute sits in telco networks, but control remains with hyperscalers; or

2) Telco-led where they host and operate edge compute, using partners such as NVIDIA as well as cloud players while controlling platform access.

Telco-led outcomes are beginning to emerge, but they vary in maturity:

• T-Mobile: working with NVIDIA and Nokia to turn its 5G network into distributed edge AI infrastructure, with pilot use cases including smart-city traffic optimization, utility inspection and industrial safety monitoring, alongside live field demonstrations of concurrent AI and RAN processing.
• SK Telecom: deploying edge AI infrastructure to support real-time inference at the network edge, particularly for enterprise and industrial use cases, while building out broader AI capabilities including NVIDIA-based GPUaaS, sovereign AI infrastructure and a planned Manufacturing AI Cloud to support digital twins and AI-driven operations.
• Verizon: packaging private 5G, private MEC, NVIDIA AI Enterprise and NIM microservices into an on-premises enterprise AI solution for real-time inference, with FanDuel cited as a customer use case in live media production.

Monetizing intelligent bandwidth

The expansion into platform models does not replace existing connectivity revenues, but layers new monetization opportunities on top and will vary in scale and execution complexity. Telco monetization in the AI era can be understood as a progression across three positions in the value chain.

Smart connectivity: At the lowest level, connectivity evolves through performance-based differentiation such as QoS tiers and network slicing, providing incremental improvements to existing ARPU-driven models. Key enablers here include traditional vendors like Ercisson and Nokia which anchor the RAN and core infrastructure. 

Network-as-a-Platform: A more structural shift comes from exposing network capabilities via APIs, enabling developers and enterprises to consume functions such as identity, location and quality of service on a usage basis. This creates a path for telcos to move beyond access pricing toward monetizing network functions embedded within applications. Industry initiatives such as GSMA Open Gateway are standardising these APIs globally, while hyperscalers and cloud platforms increasingly influence how these capabilities are integrated into developer ecosystems, creating a contested control point between telcos and cloud providers.

Edge AI provider: The largest incremental opportunity lies at the network edge, where telcos can support latency-sensitive AI workloads by combining connectivity, on or near-site compute, and AI software into integrated enterprise solutions. However, this layer remains contested, as many deployments sit on enterprise premises or metro edge infrastructure where hyperscalers retain control of the compute and developer ecosystem. Telcos have the strongest potential advantage at the RAN edge, where AI workloads can be tightly integrated with network infrastructure, though this remains at an early stage of deployment.

The applicability of these models depends on the requirements of underlying AI workloads, which determine where compute must be executed.  

AI Workload Requirements and Telco Opportunity Areas

The immediate and growing opportunity for telcos lies in latency-sensitive workloads where network proximity and real-time control create defensible competitive advantages.

The shift from pipe to platform is not universal across all AI workloads, but highly selective. While hyperscalers will continue to dominate centralized AI infrastructure, telcos have a meaningful opportunity to capture value at the edge. Success will depend less on technology deployment and more on the ability to identify and execute against the right use cases where network capabilities translate into differentiated value.