Are telcos truly slow in adopting AI? Are they really sitting on a gold mine and using it as a chair? While even the banking industry has moved fast to adopt AI, why does the telco industry lag behind? The basic adoption we see in telcos is still around network fault detection, customer service chatbots, and fraud detection. Are there any telcos doing anything different and what opportunities do they have?

Are telcos truly slow in adopting AI? Are they really sitting on a gold mine and using it as a chair?

Telcos are not bad at technology

Let us address each of these questions one by one. First, the notion that telcos are bad at technology is baseless. These companies run some of the most complex, real time distributed systems on the planet. A major carrier is managing millions of concurrent connections, routing packets across continents in milliseconds, orchestrating spectrum allocation in real time. Telco people are not dumb.

The issue is something more subtle. Telcos are bad at a very specific kind of technology: the kind that threatens to change who they are, not how they operate. And AI does exactly that.

The mental model baked into telcos for over a century is this: we own the pipe, you pay to use the pipe. Everything else is noise. The business model is subscription based access. The competitive advantage is geographic coverage and spectrum licences, neither of which you can code your way out of or into. It is how this industry was built.

Banks had the same problem for years. They thought they were in the vault business. But they eventually realised they were actually in the trust and financial flow business. The ones that figured it out early became industry leaders. The others were cautionary tales in economics and MBA programmes.

Server infrastructure representing legacy telco systems
Legacy BSS and OSS infrastructure, architected before the iPhone existed, remains a defining constraint for most major carriers.

The real blocker: legacy systems and a lack of urgency

Every conversation about telco and AI eventually hits the same wall. Our systems are old. And that is true.

Many major carriers run BSS and OSS stacks that were architected before the iPhone existed. There are transformation programmes underway and yours truly has consulted on some, but there is a long way to go. The customer record for your phone plan might live in a system that communicates via SOAP over a VPN that has not been meaningfully updated in the last decade.

You cannot just swap in a transformer model when your data is scattered across seventeen siloed systems with inconsistent schemas. Any serious AI programme at a telco starts with a data infrastructure problem. That is not an easy or cheap problem to solve.

But every industry has legacy debt. Healthcare has HIPAA constrained HL7 messages flying between systems that look like they were built in 1996. Financial services run COBOL that no one alive fully understands. I have provided consulting services to an e-commerce company whose billing was built on a system that only one person, close to retirement, knew how to configure. And these industries still found ways to move on AI.

In telcos, the existing business still works. ARPU is declining slowly, not catastrophically. There is no burning platform. So the hard infrastructure work keeps getting deprioritised to the next quarter, if not the next financial year.

17
Siloed systems in a typical carrier data estate
4
Core AI use cases deployed across most telcos
~0
Revenue generating AI plays at scale in most operators

What AI looks like in most telcos today

It is not that no AI implementation exists. Network fault detection and predictive maintenance were among the first use cases. Churn prediction is another, with propensity models flagging customers likely to leave. Still valuable, but not exactly frontier stuff.

Then there are chatbots. We know the customer service chatbots. We are also aware how mediocre they have been. Addressing actual customer issues has always seemed less of a driving factor than reducing headcount.

Fraud detection, covering SIM swap, international revenue share fraud, and account takeover, is probably the best executed AI vertical in telco. Real time scoring on call patterns and authentication signals. Here, the financial incentives were direct enough to drive genuine rigour.

  • Network operations: fault detection and predictive maintenance
  • Customer care: chatbots, often mediocre in practice
  • Retention: churn and propensity models
  • Security: fraud detection covering SIM swap, revenue share fraud, and account takeover

Do you see anything missing? There is almost nothing about building new revenue streams. Nothing about turning the network into a platform. Almost nothing about leveraging data assets in ways competitors cannot replicate. The data telcos hold is enormous: call records, location traces, network telemetry, device metadata, payment histories, and customer support logs. It genuinely frustrates me that this is not leveraged sufficiently.

There is almost nothing about building new revenue streams.

The opportunity surface: five plays hiding in plain sight

The opportunity is enormous and mostly untouched. These are not research problems. They are integration and will problems.

// 01
Hyper personalised dynamic pricing
Telcos hold extraordinary behavioural signals including usage patterns, location sequences, device types, and roaming behaviour. An AI system that dynamically constructs a personalised plan in real time is not a research problem. It is a will problem.
// 02
Selling intelligence, not connectivity
The network is a sensor grid. Every packet is a signal. Anonymised mobility data tells you where people go, at what times, in what densities. Industries like retail, urban planning, and logistics pay for survey based proxies of this data.
// 03
Sovereign AI infrastructure
Governments want AI infrastructure in country with verifiable data residency. Telcos own data centres and fibre, have regulatory relationships, and carry a century of trusted national infrastructure status that hyperscalers can never replicate.
// 04
SME AI bundling
Telcos have direct billing relationships with millions of small businesses. This is a distribution channel Microsoft would pay handsomely to own. Bundle AI tools into business plans the way Copilot is bundled into M365.
// 05
Agentic network operations
Not just alerting humans to problems but diagnosing, rerouting, scheduling engineers, ordering parts, and filing regulatory paperwork through a coordinated stack of AI agents. Telcos that get there first will have a structural cost advantage that is very hard to replicate.
Global network connectivity representing telco infrastructure opportunity
The telco asset base, covering fibre, spectrum, data centres, and edge compute, maps almost perfectly to what sovereign AI infrastructure requires.

What leading operators are doing differently

Credit should be given where it is due. Some operators are already proving this can be done at scale, not as pilots.

// ASIA
SK Telecom
Created a separate AI company inside the company specifically to route around legacy bureaucracy. AI revenue went from 9% to 22% in a year. Their personal AI assistant has 5.5 million subscribers and is evolving into a super app. They also co-founded Syntelligence AI, a joint venture with Singtel, SoftBank, Deutsche Telekom, and e&, to build telco specific LLMs across 1.3 billion subscribers.
9% to 22% AI revenue in 12 months
// WEST
T-Mobile
Built IntentCX, a platform trained on billions of actual customer interaction datapoints that connects directly to transaction systems to take autonomous action. Not a chatbot. The architecture is fundamentally different. Early rollout reached customers in 2025.
Autonomous action. Not a chatbot.
// JAPAN
NTT
Asking a prior question: what does the physical world need to look like for AI to work at the scale it is heading toward? Their IOWN initiative bets that the physical infrastructure layer is the strategic moat. All photonic networks using light instead of electronics that dramatically cut latency and power consumption for AI inference at scale.
IOWN. Photonics. Infrastructure moat.
If you were designing the perfect company to be infrastructure for the AI era, you would describe a telco.

The bottom line

The telco industry has what it needs to be the centre of AI infrastructure: nation scale sensor networks, optical fibre in the ground, data centres in the right places, regulatory trust that hyperscalers will never have, billing relationships with businesses that exist, and edge compute locations that agentic AI will increasingly depend on.

And still, most of the industry is spending time optimising EBITDA and running pilots that never become products.

Telcos have survived every platform shift since the telegraph. The internet was supposed to make them irrelevant. The smartphone was supposed to commoditise them completely. They are still here, still essential, still embedded into the fabric of how the world communicates. That is not luck. That is deep infrastructure that compounds over time.

The AI era does not diminish that. It amplifies it. Once telcos stop thinking of themselves as pipe companies and start becoming the physical layer of the AI economy, they will be extraordinarily well positioned. Not despite the infrastructure, but because of it.

// key takeaways
01 Telcos do not lack technical capability. They lack appetite for business model change.
02 Most AI deployments in telcos are cost and risk plays, not growth plays.
03 The asymmetric upside sits in platform economics: pricing, data products, distribution, and sovereign infrastructure.
04 Some operators are already proving it can be done at scale, not as pilots.