Is Your Organisation Built to Capture It?

The numbers are no longer ambiguous. Appledore Research projects $60 billion in industry savings from AI across global telecoms by 2030. NVIDIA’s 2026 State of AI in Telecommunications survey reveals that 89% of telcos plan to increase AI spending this year: up from 65% in 2024. The ITU has formalised AI’s role in network architecture. And 71% of operators are planning agentic AI deployments before the year is out.

Yet the most important story isn’t the size of the opportunity. It’s the gap between those capturing it and those still running proof-of-concepts that will never see production.

Telcos sit on some of the richest operational data on the planet (network performance, customer behaviour, fault patterns, traffic flows) and most of it remains deeply underutilised. The organisations that close this gap in the next 24 months will define the next generation of the industry. 

Those that don’t will find themselves outmanoeuvred by leaner operators, hyperscalers, and AI-native challengers who don’t carry the legacy weight.

This article examines where AI is creating the most significant telco disruption right now, what’s separating early leaders from the pack, and where Teraflow fits into the transformation backbone that makes this next chapter real.

The State of AI in Telecoms: What the Research Actually Shows

Spending Intent Is Accelerating (But Unevenly)

The NVIDIA 2026 survey is striking not just for its headline figure (89% of telcos increasing AI spend), but for what it reveals beneath the surface. In 2024, only 65% of operators had the same intent. That 24-percentage-point jump in a single year reflects a market that has decisively shifted from experimentation to investment. But investment intent and deployment maturity are very different things.

The research identifies clear pockets of high ROI: network operations, customer care automation, predictive maintenance, and revenue assurance. These are the domains where the data infrastructure exists, the use cases are well-understood, and the ROI is measurable. Elsewhere (particularly in strategic AI deployment and autonomous decision-making) most operators are still finding their footing.

Agentic AI Is the Next Frontier and It’s Arriving This Year

Perhaps the most significant finding in recent telco research is the pace of agentic AI adoption. RADCOM’s 2026 survey found that 71% of telcos plan to deploy agentic AI systems within the year. This isn’t incremental. Agentic AI (systems that can autonomously plan, act, and adapt across multi-step workflows without human intervention at each stage) represents a fundamental shift in how networks are managed and how customers are served.

The use cases showing the clearest early traction are security threat detection and response, complex customer complaint resolution, network fault diagnosis and autonomous remediation, and intelligent workforce augmentation in NOC and SOC environments. 

The organisational implications are profound: deploying agentic AI at scale requires data readiness, model governance, integration architecture, and change management capabilities that most telcos are still building.

AI-Native Networks: The Architecture Shift Happening Before 6G

77% of telcos expect AI-native network architectures to launch before full 6G rollout. This matters because it signals that AI integration is no longer a future consideration tied to next-generation standards, it’s a present-tense infrastructure decision.

AI-native networks go beyond AI-assisted operations. They embed intelligence into the network layer itself: dynamic spectrum allocation, self-healing topology, real-time traffic prediction and rerouting, and automated capacity planning. The distinction between AI-augmented and AI-native is not semantic, it determines whether AI is a tool applied to the network or a foundational principle of how the network functions.

The ITU-T TR.GenAI-Telecom framework, published in March 2025, formalises this architectural evolution and maps the intersection of generative AI with core network functions, from troubleshooting and configuration to customer interaction and service assurance. Critically, the framework also surfaces the risk dimensions: hallucination in autonomous systems, data privacy in AI-driven personalisation, and the governance requirements for deploying generative models in regulated telecommunications environments.

GenAI in Telecoms: A Fast-Growing Market With Uneven Deployment

The GenAI telecoms market is valued at $428 million in 2025 and is projected to reach $606 million in 2026, strong growth in absolute terms, but still modest relative to the scale of the industry’s overall AI investment. The concentration of GenAI deployment in customer-facing applications (conversational AI, personalised offers, churn prediction) reflects where the data pipelines and business cases were most mature.

What is shifting is the expansion of GenAI into operational domains: network documentation, automated root-cause analysis, code generation for network configuration, and knowledge management for field engineering teams. This is where the next wave of GenAI ROI will emerge and it requires a very different infrastructure foundation than customer-facing deployments.

Open-Source AI Is Winning the Infrastructure Argument

89% of telcos are prioritising open-source AI models, a figure that deserves attention from any organisation making platform decisions right now. The drivers are well understood: cost efficiency, customisation depth, avoidance of vendor lock-in, and the ability to deploy models on-premises or in private cloud environments where data sovereignty requirements apply.

This shift has significant implications for the Microsoft and Azure ecosystem that currently dominates telco AI tooling. It does not displace proprietary platforms entirely, but it does change the architecture. Operators are increasingly building modular AI stacks: open-source foundation models fine-tuned on proprietary network data, integrated with existing OSS/BSS platforms through orchestration layers, and governed through purpose-built MLOps pipelines.

What Separates the Leaders From the Rest

The research consistently points to a set of differentiating capabilities among telcos that are capturing real AI ROI: as opposed to those accumulating impressive-sounding pilot portfolios.

• Data as the Foundation: A unified, accessible data layer. Operators like Vodafone and AT&T, which have publicly reported 10x agility improvements and 50% reductions in customer query handling time respectively, share a common characteristic: they invested early in data infrastructure. Not AI infrastructure, data infrastructure. Clean, governed, well-modelled operational and customer data is the prerequisite for every downstream AI use case.
• Production-Grade ML: ML in production, not proof-of-concept. The gap between a model that performs in a sandbox and one that delivers value in a live network environment is enormous. Leaders have built the MLOps discipline to bridge this gap: model monitoring, retraining pipelines, performance tracking, and rapid iteration loops.
• Software Architecture: Modern, API-first software architecture. AI use cases don’t exist in isolation. They need to integrate with network management systems, customer platforms, billing, workforce management, and third-party ecosystems. Operators with modular, API-driven architecture can deploy and scale AI capability orders of magnitude faster than those relying on monolithic legacy stacks.
• Deployment Methodology: Speed-to-value deployment methodology. The operators demonstrating measurable ROI are not running 18-month transformation programmes. They are running focused, time-boxed deployments that put AI into production in weeks, measure outcomes, and iterate. The methodology matters as much as the technology.

Where Teraflow Fits: Building the Transformation Backbone

Teraflow was built for precisely this inflection point. Not to sell AI strategy, there is no shortage of strategy in this industry, but to deliver the practical infrastructure and deployment capability that converts AI investment into operational reality.

Our work with enterprise telecoms organisations is structured around four interdependent pillars that address the specific capability gaps the research identifies as most consequential.

Data: The Foundation Everything Else Depends On

Every significant AI deployment in telecoms (network optimisation, customer intelligence, predictive maintenance, autonomous fault resolution) is downstream of data quality. Teraflow’s data engineering practice helps operators design and implement the data foundations that make AI deployment viable at scale.

This means data architecture that reflects the real complexity of telecoms operations: OSS and BSS integration, network telemetry pipelines, customer data unification, and the governance frameworks that ensure data can be used confidently in automated systems. We don’t start with the model, we start with the data that will make the model useful.

ML: From Experimentation to Production Value

The industry’s biggest AI ROI gap isn’t between operators with AI ambitions and those without. It’s between organisations that can get models into production (reliably, repeatably, at scale) and those that can’t.

Teraflow’s machine learning engineering capability spans the full production lifecycle: use case prioritisation, model development, feature engineering, evaluation frameworks, MLOps infrastructure, and ongoing performance management. We bring particular depth in the use cases the research identifies as highest-value: network anomaly detection, churn and lifetime value prediction, demand forecasting, and intelligent automation for network operations.

Software: The Integration Layer That Makes AI Real

AI capability that can’t integrate with existing operational systems doesn’t deliver value, it delivers demos. Teraflow’s software engineering practice builds the integration fabric that connects AI models to the systems they need to act on: network management platforms, customer databases, ticketing systems, workforce management tools, and the emerging agentic orchestration layers that coordinate multi-model workflows.

With 89% of telcos prioritising open-source AI models, the architecture decisions being made right now will shape operator capability for the next decade. Teraflow helps operators make these decisions with clarity, designing modular, maintainable systems that avoid lock-in while delivering the integration depth that production AI requires.

DAPA: 30 Days to Measurable Value

Teraflow’s proprietary DAPA methodology is our response to the deployment speed gap that separates AI leaders from the rest of the market.

DAPA is designed to take a telco organisation from a defined business problem to a live, value-generating AI deployment in under 6 months, not years. It is structured around the principle that speed-to-production is itself a strategic capability and that organisations which learn to deploy AI rapidly build compound advantages over time.

The Next 24 Months: What Telcos Should Be Acting On Now

The research is directionally consistent across every major source: AI investment in telecoms is accelerating, agentic AI is moving from aspiration to deployment, and the organisations that establish production AI capability in the next two years will have structural advantages that are very difficult to overcome later.

Based on the current state of the market, we see five specific areas where action in the next 24 months will prove most consequential.

• Prioritise data foundations first: Invest in data infrastructure before AI infrastructure. The ROI gap in most organisations is not a model problem, it is a data problem. Treat data engineering as a first-order strategic investment, not a prerequisite to be solved later.
• Deploy to production, not proof-of-concept: Identify two or three high-value, high-data-readiness use cases and deploy them to production using a rapid methodology. Measurable production deployments build more organisational confidence and capability than ten parallel pilots.
• Build the governance and architecture for agentic AI: Understand what agentic AI actually requires (data readiness, governance frameworks, integration architecture, human oversight protocols) and begin building those foundations now, ahead of the deployment wave.
• Make platform decisions with long-term architecture in mind: 89% open-source adoption is not a trend to wait out. Make deliberate platform decisions now that preserve flexibility and avoid the lock-in that will constrain your options as the model landscape continues to evolve rapidly.
• Build internal AI capability in parallel with deployment: AI deployment at scale requires internal capability, not just external partnerships. The operators building internal ML engineering, data science, and AI governance competency are creating durable advantages. Capability transfer should be a feature of every external AI engagement.

Closing Perspective

$60 billion in industry savings is not a forecast to file away. It is a redistribution, a reallocation of value from operators who move slowly to those who move with purpose. The technology is no longer the constraint. The constraint is the ability to execute: to build the data foundations, deploy production-grade ML, integrate AI into operational systems, and do it at a pace that creates real competitive distance.

This is the work Teraflow was built for. Not transformation theatre.

Not roadmaps that age in a drawer. Practical, production-grade AI deployment that puts capability into the hands of the people running the network and creates measurable value in weeks, not years.