AI in Telecom: 5 Key Takeaways for Operators
Artificial intelligence is becoming deeply embedded in telecom operations, but the industry’s largest AI opportunity may not be building or selling foundation models.
A recent TelecomTV panel featuring Warren Bayek, Vice President of Intelligent Edge, Software and Services at Aptiv; Robert Curran, Consulting Analyst at Appledore Research; and Diego R. Lopez, Senior Technology Expert at Telefónica and chair of ETSI ZSM ISG and NOC of ETSI ISG NFV, explored where AI is creating practical value—and where the industry’s expectations still exceed operational reality.
Telecom operators already use machine learning for network anomaly detection, predictive maintenance, traffic optimization, customer-care automation, and AIOps. The next phase, however, introduces agentic AI: systems capable of interpreting objectives, invoking tools, coordinating workflows, and potentially taking actions across operational domains.
That transition creates new requirements around trust, governance, interoperability, sovereignty, and operational safety.
The panel also highlighted a potentially important strategic position for telecom operators. Rather than competing directly with hyperscalers and foundation-model providers, operators can leverage assets they already control: distributed infrastructure, network intelligence, connectivity, edge locations, regulated environments, and operational expertise.
Here are five key takeaways.
🤖 Agentic AI Is Advancing, but Trust Remains the Constraint #
Telecom operators have used AI and machine learning for years. Network anomaly detection, predictive maintenance, traffic optimization, customer support automation, and AIOps are established use cases.
Agentic AI changes the operating model.
Instead of producing only predictions or recommendations, an AI agent can interpret a goal, select appropriate tools, execute a workflow, evaluate results, and potentially take further actions. In a telecom environment, that could mean coordinating activities across network management, service assurance, IT operations, customer systems, and other operational platforms.
The distinction is significant because telecom infrastructure is highly heterogeneous.
Lopez argued that agentic AI remains relatively immature for mission-critical telecom environments, particularly because operators depend on diverse infrastructure and vendor-specific systems. Bayek likewise identified operational trust as a major barrier. Telecom operators have worked toward closed-loop automation for years, but giving autonomous systems greater control over critical services requires much stronger guarantees about their behavior.
The near-term opportunity may therefore be cross-domain orchestration rather than fully autonomous network control.
Instead of replacing existing operational tools, AI agents could coordinate them. This approach could improve operating efficiency, workforce productivity, customer experience, and the industry’s progression toward autonomous networks.
Organizations including GSMA, TM Forum, and Microsoft are already exploring agentic AI architectures, governance, standards, and interoperability for telecom environments.
The core engineering challenge is consequently expanding beyond model performance. It increasingly involves system integration, observability, authorization, policy enforcement, deterministic safeguards, and governance.
🌐 Telecom’s Biggest AI Opportunity May Be Infrastructure #
One of the panel’s central observations was that telecom operators are unlikely to become major foundation-model providers.
Instead, operators can potentially position themselves as infrastructure providers and trusted intermediaries for AI workloads.
Lopez described telecom infrastructure as a potential “substrate” for AI, analogous to the industry’s role in IoT. Bayek highlighted several assets that distinguish telecom operators from conventional cloud providers:
- Distributed infrastructure
- Extensive network footprints
- Local and regional facilities
- Regulatory and compliance expertise
- Secure operational environments
- Existing relationships with enterprises and governments
- Low-latency connectivity
This creates an alternative path to telecom AI monetization.
Rather than competing directly with hyperscalers and model developers, operators can provide infrastructure for deploying AI closer to users, enterprises, devices, and regulated data.
Potential services include AI hosting, accelerated computing, edge inference, AI APIs, distributed cloud platforms, and infrastructure services.
The strategic advantage is not necessarily the underlying AI model. It is the combination of compute, connectivity, geographic distribution, network intelligence, and operational control.
💰 AI Revenue Is Still Unproven, but Monetization Is Expanding #
Telecom operators have experienced ambitious technology-revenue forecasts before, particularly during the 5G investment cycle. Curran noted that some of AI’s earliest financial benefits may come from preventing revenue leakage, reducing churn, and improving operational efficiency rather than creating entirely new revenue streams.
AI is already being applied to areas such as:
- Customer-care automation
- AIOps and network operations
- Predictive maintenance
- Network optimization
- Churn reduction
- Service assurance
- Workforce productivity
At the same time, operators are exploring direct AI-related revenue opportunities, including:
- AI hosting
- GPU-as-a-Service
- Edge inference
- Enterprise AI platforms
- Data services
- AI infrastructure
Bayek also pointed to physical AI and edge AI as areas where operators could participate without attempting to compete in foundation-model development.
The distinction between AI-driven efficiency and AI-generated revenue is important. The former can often be measured through reduced costs or improved operational metrics, while the latter requires customers to pay for new AI-enabled products and infrastructure.
A larger opportunity could emerge as network infrastructure and AI infrastructure converge.
Distributed compute resources could potentially support multiple workloads, including RAN functions, network applications, enterprise workloads, and AI inference. Better infrastructure utilization could make AI services economically relevant to telecom assets that historically existed primarily to operate communications networks.
🛡️ Sovereign AI Could Become a Strategic Requirement #
Sovereignty was another major theme in the discussion.
Curran framed the AI-factory conversation around a broader question: who controls the infrastructure, data, models, and operating environment supporting critical AI workloads?
For governments and regulated industries, AI deployments can involve requirements related to:
- Data residency
- Regulatory compliance
- Local operational control
- Cybersecurity
- Supply-chain resilience
- Geopolitical risk
- Continuity of critical services
These requirements can create demand for sovereign AI infrastructure operated within specific jurisdictions rather than relying exclusively on globally distributed hyperscale platforms.
The panel therefore discussed the possibility of smaller, targeted sovereign AI facilities designed for regional and regulated workloads.
For telecom operators, sovereignty can represent more than a compliance requirement. Their existing infrastructure, local presence, security capabilities, and relationships with governments and enterprises can support AI environments where customers require tighter control over data and operations.
This also connects with the broader evolution of sovereign cloud and trusted AI infrastructure.
From an architecture perspective, sovereign AI can involve constraints across the entire stack—not just where data is stored. Compute infrastructure, model hosting, identity, network connectivity, management planes, logging, security controls, and operational personnel may all become relevant to sovereignty requirements.
⚡ Edge AI Could Strengthen the Business Case for Telecom Edge Computing #
The panelists identified edge AI as a natural opportunity for telecom operators.
The underlying proposition is straightforward: operators already control distributed infrastructure, network edge locations, connectivity, and facilities positioned close to users and devices.
AI workloads that require low latency, local processing, data privacy, or real-time decision-making can benefit from this footprint.
Potential applications include:
- Industrial automation
- Robotics
- Computer vision
- Physical AI
- Security systems
- Real-time enterprise applications
- AI-assisted network operations
- Mission-critical services
Bayek’s observation that “only telcos own the edge” captures an important distinction. The potential advantage is not simply access to edge compute. Telecom operators combine compute locations with connectivity, network intelligence, physical infrastructure, and proximity to devices.
The difficult part is turning that technical capability into commercially viable services.
Edge AI could therefore connect two previously separate technology discussions: the evolution of telecom edge computing and the growing requirement for distributed AI inference.
Architecturally, this also shifts some AI workloads away from centralized data centers. Inference can be placed closer to data sources and end users when latency, bandwidth, privacy, or availability requirements justify the additional operational complexity.
🚀 AI in Telecom Is Becoming an Infrastructure Story #
The panel discussion points toward a broader shift in how the telecom industry’s AI opportunity can be understood.
Operators do not necessarily need to compete with hyperscalers on foundation models to participate in the AI economy. Their existing infrastructure can provide value in areas where distribution, connectivity, trust, sovereignty, and proximity matter.
Three themes stand out:
- Automate network operations: Apply AI to AIOps, network automation, service assurance, and progressively more autonomous operational workflows.
- Monetize distributed infrastructure: Use edge locations, connectivity, accelerated compute, and distributed cloud infrastructure to support AI workloads.
- Build trusted AI environments: Provide sovereign and regulated infrastructure for enterprises, governments, and other customers with strict requirements around data and operational control.
The important change is conceptual: telecom AI is no longer only about deploying AI inside the network. It is increasingly about making the network itself part of the AI infrastructure stack.
If this model develops successfully, telecom operators can participate across multiple layers of the AI value chain—including infrastructure, edge computing, trusted environments, data, connectivity, and operational intelligence—without necessarily becoming AI model companies.