After a year of strategic planning and positioning, 2025 has become the year of execution, where AI moves from boardroom presentations to real-world deployments that reshape how telcos operate, compete, and create value.
According to GSMA Intelligence‘s latest market analysis covering 250 operators across more than 100 countries, the picture is becoming remarkably clear: telcos aren’t just experimenting with AI anymore.
They’re deploying it strategically, methodically, and with increasingly sophisticated understanding of where it delivers the most impact.
At Teraflow, we’re helping this transformation unfold in real-time, and the patterns emerging tell a compelling story about the future of telecommunications.
Let’s dive into the three critical areas where telcos are concentrating their AI efforts (and why these choices matter for the industry’s next chapter).
1. Customer Care: The Low-Hanging Fruit That’s Bearing Real Results
(The report reveals a standout 47% in customer care AI deployments, with further key stats highlighting the growing investment across the sector. Source: GSMA Intelligence, “Telco AI: State of the Market, Q2 2025”)
The Numbers Tell the Story
Customer care has emerged as the undisputed leader in telco AI deployments, accounting for nearly half (47%) of all tracked implementations. This isn’t surprising, it’s where the business case is clearest, the ROI is fastest, and the technology is most mature.
But here’s what makes this trend particularly interesting: around 75% of AI deployments in customer care are already live, making it the most mature category in the telco AI landscape.
These aren’t pilot programs or proof-of-concepts anymore. They’re full-on operational systems handling millions of customer interactions every day.
Why Customer Care First?
The answer comes down to three words: automation, efficiency, and scale.
Customer care deployments, driven by cost savings from automation, also support churn reduction and upsell objectives. It’s the rare use case where cost reduction and revenue enhancement can coexist harmoniously.
Think about it from a telco’s perspective: call centers represent a massive operational expense, with agents handling repetitive queries that AI can manage with increasing sophistication.
AI agents and chatbots can now resolve routine issues (password resets, bill inquiries, plan changes) while escalating complex problems to human agents who can focus on what they do best: building relationships and solving nuanced challenges.
Recent developments underscore this momentum: Ooredoo has unveiled a GPT-4o powered AI chatbot called “Obot” across key customer touchpoints, while A1 and Cognigy are bringing AI into corporate customer service.
These aren’t isolated experiments: they’re strategic bets on AI as the foundation of modern customer engagement.
The Regional Reality
The focus on customer care isn’t uniform globally, but it’s universal.
In Latin America, customer care represents 67% of AI deployments, while in MENA it accounts for 33%. Even at the lower end, it’s still the single largest category, reflecting both the universal applicability of the technology and the pressing need for operational efficiency in a low-growth environment.
What excites us is watching these deployments evolve beyond simple chatbots into intelligent systems that understand context, anticipate needs, and deliver genuinely personalised experiences at scale. Very much like our approach to AI.
2. Network Operations: Where AI Becomes Mission-Critical Infrastructure
The Strategic Imperative
While customer care grabs headlines, networks represent where AI moves from efficiency tool to strategic differentiator.
Networks account for almost 20% of telco AI deployments, and this number likely understates reality: many network AI initiatives remain confidential for competitive reasons.
Here’s the critical insight: in reality, almost all operators will use AI in their networks, but there is often overlap with other functions. AI in networks isn’t a single use case, it’s a constellation of applications that touch everything from predictive maintenance to energy optimisation to performance enhancement.
From Reactive to Predictive
Traditional network management has always been fundamentally reactive: problems occur, alarms fire, engineers respond.
AI flips this model entirely, enabling pre-emptive fault detection that catches issues before customers notice them. This isn’t just about avoiding complaints: it’s about fundamentally reimagining network reliability.
The technology front is evolving rapidly.
Ericsson and Bell Canada have successfully tested AI-native link adaptation to boost network speed and efficiency, while Ericsson and Google Cloud are teaming up to deliver carrier-grade 5G core as-a-service built with AI at the foundation.
These aren’t incremental improvements, they’re architectural shifts that embed intelligence directly into network infrastructure.
All major equipment vendors have signaled support for deeper AI integration, though greater quantification of AI benefits in individual network categories would be beneficial.
This points to where the industry is heading: AI-native network architecture where intelligence isn’t bolted on, but built in from the ground up.
The Energy Angle
One often-overlooked network AI application deserves special attention: energy optimisation.
With data traffic continuing its relentless growth and energy costs representing a significant operational expense, AI-driven energy management through intelligent sleep states and dynamic resource allocation isn’t just environmentally responsible, it’s financially essential.
3. Edge Inference: The Emerging Battleground for Revenue Growth
The Paradigm Shift
If customer care and network operations represent AI’s early wins, edge inference represents its future. This is where the conversation shifts from cost savings to revenue generation, from internal efficiency to external innovation.
The value of the edge comes down to several benefits, including cost savings (backhaul, storage and egress), productivity gains, data sovereignty and resilience.
But here’s what makes edge inference particularly compelling: running inference at the edge has several selling points for operators compared to processing AI workloads in the cloud, with cost savings potentially reaching 30-40%.
Training in the Cloud, Inference at the Edge
Understanding the edge inference opportunity requires recognising the fundamental division of labor in AI infrastructure.
The public cloud provides the workhorse compute power behind most LLM training, while the edge is where much of the AI inference takes place in the operator context.
This architectural split creates a strategic opening for telcos.
While hyperscalers dominate training infrastructure, telcos control the edge – the physical infrastructure closest to where data is generated and decisions need to be made in real-time.
Where the Action Is
Early inference use cases for operators concentrate on on-premises deployments for a range of enterprise segments, including digital twins, robotics, and industrial IoT.
Think hospital floors with AI-powered inventory management, factory floors with predictive maintenance, retail environments with intelligent security and customer analytics.
Real-world momentum is building. Safaricom and iXAfrica Data Centres have entered a strategic partnership to deliver Kenya’s first AI-ready infrastructure for enterprise innovation, while Indosat Ooredoo Hutchison has inaugurated an AI experience center in Jayapura, bringing tangible AI benefits to Eastern Indonesia.
The Revenue Challenge
Here’s the reality check: only 10-20% of telco AI deployments are in place to drive revenues, with the rest primarily targeting internal efficiencies.
This reflects both the maturity curve (cost savings are easier to quantify and deliver) and the reality that many AI-driven revenue models are still emerging, including GPUaaS and agentic AI.
The challenge is less in the technology and more in the revenue model, as this is quickly becoming a crowded field. Telcos aren’t the only ones eyeing edge infrastructure: cloud providers, CDN operators, and specialised edge computing companies are all competing for the same enterprise customers.
But this is exactly where opportunity lies.
Telcos have unique advantages: national-scale infrastructure, existing enterprise relationships, regulatory compliance expertise, and the trust that comes with being critical communications providers.
The question isn’t whether edge inference will be significant, it’s which players will capture the value.
The Bigger Picture: A Long-Term Transformation
Cycles Within Cycles
One of the most important insights from the GSMA Intelligence analysis is this: 60% of tracked AI deployments have already been launched by operators as part of their day-to-day business, with the remaining 40% in trial or planning stages.
But here’s what makes AI different from previous technology waves: unlike 3G/4G/5G networks, where trials linearly give way to live networks over a 10-year cycle, AI is likely to be a repeated pattern of ‘cycles within a cycle’, with trials representing a large proportion of total deployments.
This isn’t a one-and-done upgrade cycle. It’s continuous innovation, with operators constantly testing, validating, deploying, and iterating.
2025 is a year of transition from trials and validation to commercialization across all regions, with the next two to three years seeing repeating cycles of test, validate, deploy and innovate.
The Upskilling Imperative
Technology is only part of the equation. Across many AI investment requirements, 63% of operators prioritized upskilling, closely followed by network capacity at 62%.
This upskilling priority reflects the reality that AI isn’t just a technology you deploy, it’s a capability you must continuously develop and refine.
At Teraflow.ai, we see this playing out in real-time. The telcos succeeding with AI aren’t just buying platforms, they’re building expertise, fostering experimentation, and creating cultures where learning and adaptation are embedded in the DNA.
Sovereign AI: The Strategic Wildcard
One theme deserving special attention is sovereign AI.
Sovereign AI is likely to become an even stronger current in 2025, representing a key competitive advantage for operators as communication infrastructure players on a national scale.
Recent developments underscore this trend. Saudi Arabia and Nvidia are building AI factories to power the next wave of intelligence for the age of reasoning, while the European Broadcasting Union and Nvidia are partnering on sovereign AI to support public broadcasters.
For telcos, sovereign AI represents both opportunity and responsibility: the chance to position themselves as trusted national infrastructure providers for the AI age, with all the commercial and strategic benefits that entails.
Looking Forward: From Cost Savings to Value Creation
The current state of telco AI is clear: 75-80% of AI deployments have a primary motivation to save money for operators, driven by AI agents in call centers and pre-emptive fault detection in networks.
This focus on efficiency is logical, necessary, and delivering real results.
But the future is equally clear: the next wave of telco AI will be about value creation, not just cost reduction. This will change during 2025 and 2026, as more trials give way to new products.
The operators who thrive won’t be those who simply deploy AI faster, they’ll be those who deploy it smarter, with clear understanding of where AI creates value, how to capture that value, and how to continuously evolve as the technology and market mature.
At Teraflow, we’re building for this future.
Creating the infrastructure and intelligence that will power the next generation of telecommunications. Because in the end, AI isn’t about replacing what telcos do. It’s about amplifying their capabilities, expanding their possibilities, and positioning them for a future where connectivity and intelligence are inseparable.
The race isn’t to deployment, it’s to value. And it’s just getting started.
