AI at the telco edge: where to play, how to win

Deciphering the technical, monetization & partnership dilemmas

As AI reshapes telecom strategy, leading C-suite telco executives understand their best opportunities lie in directed focus on the operator and network edge infrastructure layers. Telcos must now decipher the technical dilemma (how deep to invest in edge AI infrastructure), the monetization dilemma (who pays for the edge and how), and the partnership dilemma (where to compete, where to partner) to determine whether they can become critical enablers of the AI economy.

THE EDGE IS MORE CRITICAL THAN EVER

Back in 2021, Arthur D. Little (ADL) argued that edge computing was not just a technical evolution, but a strategic inflection point for telcos (see ADL report “Edge Computing: Hype or Ripe?”). At the time, edge remained optional; today, AI workloads are turning it into a race for relevance (see Figure 1). Emerging use cases such as drone logistics, swarm robotics, and AR/VR require ultra-low latency and real-time inference that cannot be fully performed on-device due to power and compute constraints.

Although AI dominates telco agendas, many operators lack clear monetization strategies and deployment timelines. Unlike video streaming, where content delivery networks and localized content caching solved bandwidth and caching challenges, AI inference is interactive and highly latency-sensitive, requiring processing close to the user. Even small rendering delays can undermine commercial viability. This means networks must evolve into distributed compute fabrics. As inference costs decline and AI applications scale, edge infrastructure becomes ever more critical to delivering real-time, context-aware intelligence.

To support AI-native services’ strict latency and coordination requirements, compute will quickly move from centralized clouds to the edge. This shift transforms the edge from a connectivity hand-off point into an intelligent processing layer, where edge sites become the backbone for real-time, low-latency, high-reliability services close to users and devices.

(RE)DEFINING THE EDGE

Figure 2 outlines the roles of each edge layer and highlights where telcos can realistically differentiate.

Introducing an intermediate operator edge helps capture the role of regional and metro data centers under carrier control. The network edge offers strong low-latency performance and scalable coverage but involves high complexity and cost. The local edge enables specialized enterprise processing but remains fragmented and contested. Device edge delivers the lowest latency but is largely controlled by OEMs and chip vendors, with additional device constraints limiting scalability.

The operator and network edge layers provide telcos with the greatest leverage, letting them use existing infrastructure, spectrum, and operational capabilities to deliver AI services at scale and build defensible positions. As illustrated in Figure 3, the indicated strengths of each layer determine the use cases they are best fit to support.

The value of AI at the edge is most tangible when examined through the lens of specific workloads across layers:

• At the cloud level, the focus is on large-scale AI training, recommendation engines, and non-time-critical inferencing, best suited for workloads requiring extensibility rather than low latency.
• The operator edge, anchored in metro and aggregation sites, enables low-latency B2B2X services.
• At the network edge, colocated with RAN or aggregation routers, telcos can leverage AI for RAN optimization. The exact count of operator and network edge can depend on operator topology (metro versus regional core), latency targets, and whether compute is placed at aggregation points of presence (PoPs) or extended toward access layers.
• The local edge, typically enterprise-controlled or hyperscaler-controlled, addresses industrial automation, healthcare, and retail use cases, where telcos often act as managed service providers.
• The device edge brings AI inferencing directly into end-user devices such as Internet of Things (IoT) sensors.

THREE TELCO DILEMMAS

The edge is not an open frontier; it has been shaped by powerful incumbents across layers. For telcos, the critical question is not whether to play, but where to play. Each layer comes with a distinct mix of demand and defensibility:

• At the cloud level, hyperscalers dominate through scale, developer ecosystems, and R&D velocity. Telcos have little defensibility here, with partnerships the only realistic option.
• The device edge is similarly out of reach; it’s owned by OEMs and chipset vendors, leaving telcos with marginal orchestration roles.
• At the local edge, enterprises and hyperscalers control much of the real estate and application stack. Telcos can participate through partnerships and integration services, but not as sole owners of the value chain.
• The operator edge and network edge are telcos’ strongest positions. The operator edge, tied to metro sites and aggregation sites, provides the right mix of scale, proximity, and defensibility. The network edge is primarily telco-internal. These provide tangible OPEX savings, though limited external monetization potential.

Three dilemmas emerge for telco CTOs, COOs, and CFOs:

1. Technical — how deep to invest in edge AI infrastructure? For AI workloads at the operator and network edge, accelerators such as GPUs (or alternatives like NPUs) are often needed to deliver real-time inference. The decision is not whether GPUs are useful (they clearly are), but where to deploy them and at what density.
   • CTO lens. Should we selectively deploy GPUs only in high-demand sites or build them more widely to attract developers and use cases?
   • CFO lens. How do we balance the CAPEX and OPEX of GPU clusters against the risk of underutilization, especially when demand for operator edge AI remains uncertain?
   • COO lens. How do we operationalize GPU-enabled edge sites at scale without exploding the costs and complexity?
2. Monetization — who pays for the edge, and how? Telcos may have undercapitalized on the data center and cloud wave, but edge opportunity signals a more decisive shift. Telcos must decide whether their future positioning should remain as a dependable infrastructure provider, monetizing footprint and bandwidth even if it means being invisible in the value chain, or move up the edge stack. This decision should be made with the broader value chain in mind. If the focus is on internal efficiency, a key factor will be the ability of network equipment providers to integrate generative AI into their products. This represents a significant leap in terms of third-party integration, security, and their capacity to effectively deliver Open RAN to the market. In this scenario, it will be important to closely monitor the progress of NEPs in these areas before making a decision.
3. Partnership — where to compete, where to partner? Telcos cannot realistically out-invest hyperscalers in GPUs or developer ecosystems. Strategic partnerships with technology players should allow telcos to offer edge AI without carrying the full burden of infrastructure developments. But partnership involves sharing value and ceding some control, raising the question of whether telcos will remain infrastructure providers or become co-creators or enablers.

In essence:

• For CTOs, the critical choice is about architecture — how far distribution should extend and how partnerships are structured
• For CFOs, the critical choice is about capital allocation — whether AI at the edge should be seen as a cost-saving lever or as a speculative growth play, with CFOs depending on the commercial go-to-market (GTM) assessments to triangulate adoption curves and payback confidence
• For COOs, the critical choice is about operationalizing at scale — scaling edge sites without getting overwhelmed by complexity or cash requirements

Furthermore, each C-suite executive must address the partnership dilemma, which will determine whether a telco remains a host inside another company’s ecosystem or fight to orchestrate one of their own.

DECISION SPACES FOR CTOs, CFOs & COOs

For telcos, AI at the edge is not a technological trend; it’s a series of strategic forks in the road. Figures 4-6 show the options, pros/cons, and C-suite takeaways for each of the three dilemmas.

Technical dilemma market insights

• Selective GPU deployment. This emerged as a pragmatic telco play, largely catering to enterprise demand through GPU as a service (GPUaaS) or GPU-powered IaaS/AI cloud instances. In the US, Verizon has introduced offerings aimed at enterprise workloads at the edge.
• Broad GPU deployment. This represents the “bet big” path, with operators investing in regional or national-scale AI infrastructure. For example, Telefónica is piloting deployment of Spain-wide distributed edge infrastructure. This strategy maximizes long-term positioning but comes with heavy CAPEX commitments and a high risk of asset obsolescence.

Monetization dilemma market insights

• B2B2X. B2B2X enablement offers telcos a lower-risk monetization path. Vodafone’s collaboration with AWS is a good example: Vodafone provides edge connectivity at its network sites while AWS supplies the compute infrastructure. This setup enables advanced enterprise use cases such as extended reality in logistics and agriculture.
• Vertical solution. Some operators pursue deeper differentiation via vertical use cases. For instance, NTT Data, a subsidiary of NTT, has developed specific edge solutions that converge IT and OT environments to deliver real-time insights.
• Internal efficiency. The most pragmatic entry point involves using AI at the edge to optimize internal operations. This includes network automation, energy optimization, and AI-driven orchestration. Examples include in-house developments like Rakuten’s RIC (RAN intelligent controller) platforms, an internal efficiency play using AI-enabled orchestration to dynamically control edge behavior and optimize network performance.

Partnership dilemma market insights

• Established vendors. Players such as Ericsson and Nokia remain centered around RAN performance and network orchestration. Offerings like Nokia’s MantaRay SMO and Ericsson’s Intelligent Automation Platform (EIAP) emphasize AI-driven optimization of network performance and energy efficiency. This reflects their core narrative: AI should be natively embedded in RAN functions to enhance efficiency and reliability.
• New entrants. Nvidia positions itself around AI workload management GPUs. Its portfolio is designed to host, orchestrate, and scale AI workloads on top of optimized connectivity, essentially promoting telcos as distributed AI compute platforms at the edge.
• Two pathways. One path for telcos involves close alignment with established vendors to extract cost efficiencies and incremental performance gains. The other involves embracing partnerships with new entrants to capture potential revenue growth.

RECOMMENDATIONS

A cautious, partnership-led approach is the most credible path for telcos in edge AI. This means focusing on use cases with clear structural advantage and demand, such as mobile AI inference for moving assets that cannot rely fully on on-device processing due to battery or compute limits. Here, network slicing could help telcos differentiate by enabling assured connectivity and quality of service (QoS) for real-time intelligence.

Selective GPU deployment may be justified at limited sites, but only where utilization, customer commitment, or operational benefits are proven. The priority should be preserving strategic option value rather than assuming broad-based monetization.

Technology lens (CTOs)

• Adopt a phased and selective GPU deployment model — start with small-volume, high-demand sites and internal efficiency use cases while designing an architecture that can scale
• Focus differentiation on telco-native domains — for example, low-latency hosting and QoS assurance rather than competing head-on in generic compute, where hyperscalers dominate
• Build selective (strategic) partnerships early to fill gaps — for example, Nvidia for AI infrastructure, hyperscalers for workload ecosystems, and vertical specialists for use-case tailoring, but retain control of telco-specific workloads, always keeping your “sweet spot” edge layers in mind

Financial & strategic lens (CFOs)

• Treat edge AI as a constrained option and not as a core growth thesis — position early GPU and edge rollouts as option bets, guided by commercial GTM assessments of pipeline and willingness to pay
• Monetize internal efficiencies first — prioritize AI for network automation and energy optimization to generate quick cost savings that can fund external edge ventures
• Negotiate partnerships for financial flexibility and limit downside — turn CAPEX into OPEX where possible (e.g., GPUaaS) but structure agreements to capture upside as demand scales

Execution & operations lens (COOs)

• Avoid scaling ahead of operational proof points — standardize site designs, automate provisioning, and build a governance model, such that selective rollouts don’t turn into bespoke deployments
• Run edge like a product — define a narrow service catalog, implement end-to-end observability, and industrialize incident playbooks across deployments to protect the customer experience
• Make the partnerships operable, not just strategic — define who does what in provisioning and incident response, aligning tooling and escalation paths to behave as a single ops team

A defensive stance may preserve CAPEX but risks reducing telcos to connectivity providers; aggressive GPU rollouts create high cost, obsolescence, and monetization risks. For more ambitious strategies, telcos should prioritize phased GPU deployments, anchor tenants, and risk-sharing partnerships to limit exposure. Investments must remain demand-backed rather than speculative. To sustain this credibly, operators need repeatable evidence and disciplined governance. Since GPU use varies widely by workload, telcos should apply clear pilot thresholds and track occupancy, per-inference cost, uptime, and QoS adherence.

Conclusion

BECOMING ESSENTIAL

AI at the edge sits at the intersection of network and compute, putting telcos in a strategic squeeze that requires clear monetization strategies and deployment timelines. To begin, they need an infrastructure plan, an economic strategy, and a strong grasp of partnership opportunities. Telcos must then decide:

1. How deep to invest in edge AI infrastructure
2. Whether to monetize through efficiency or growth
3. How to balance sovereignty with lower-risk partnerships

Success will be based on monetizing efficiency selectively in the short term, investing in focused growth, and forming partnerships that preserve strategic flexibility. Those who do so will become essential players in the AI-driven digital economy.

By Dr. Karim Taga, Christoph Uferer, Manuel Sampedro, Matyas Weber, Jatin Jain

• By Arthur D. Little
• 14/07/2026
• AI and telecom industry