Hello,

This week, telecom AI stopped looking like a collection of pilots and started looking like operating architecture.

Across Europe, Africa, North America, and Latin America, the same signal kept appearing: operators and vendors now want AI that can work across OSS, BSS, CRM, network APIs, and autonomous operations, not another isolated model bolted onto one workflow.

In today's edition:

Let’s dive in.

A 31 March analysis from Bounteous argued that telecom service quality now depends on how well ticketing systems, OSS, BSS, and CRM coordinate decisions in real time, because delays and repeat work stack up when those systems drift apart.
The same piece said the core problem is no longer that these systems are old, but that they were never designed to coordinate multi-step decisions across channels and platforms at current speed.
It also cited Gartner’s estimate that, through 2027, 50 percent of enterprises will fail to realise expected AI value because of poor integration and coordination across systems.

That argument lines up with what Caretta Research saw at MWC 2026 on 26 March.

Caretta said vendor messaging had shifted from rule-based automation to agentic BSS/OSS, with Amdocs, Netcracker, Nokia, and Ericsson pushing that direction, while Hansen, Cerillion, and Tecnotree demonstrated automation-heavy implementations.

Caretta also warned that a more autonomous platform may reduce the number of people needed for routine work, even as the specialists required to design and monitor those platforms become more expensive.

That matters because the centre of gravity in telecom AI is moving away from model novelty and towards orchestration.

A telco will not get much value from a clever assistant if the assistant cannot trigger fulfilment, pricing, assurance, and care actions across legacy estates.
The real race now is to decide which company owns the operating layer that links those workflows together.

What this means for you: map every hand-off between customer care, charging, order management, and assurance before you buy another AI product.
If those hand-offs are still manual, your next gain will come from workflow coordination, not from a larger model.

On 2 March, GSMA launched Open Telco AI, a global initiative meant to accelerate telco-grade AI through open collaboration across operators, vendors, AI developers, and academic institutions.

GSMA said the launch included a portal for telco open models, data, compute, and tools designed to speed up development and evaluation of telecom-focused AI models.

TelecomTV’s coverage the same day summed up the problem cleanly: “AI does not yet speak telco.”

TelecomTV also reported that AT&T, AMD, and TensorWave were early contributors, and quoted AT&T’s chief data and AI officer saying telecom needs models that understand real network conditions rather than generic systems repurposed for telco tasks.

AT&T said its work in the initiative is aimed at building datasets, models, and evaluation frameworks that can make generative and agentic AI useful at scale for operators.

That is a bigger deal than another MWC announcement because it points to a missing foundation layer: shared telecom-specific benchmarks, data structures, and testing environments.

Nokia’s 3 March announcement about expanding its Network as Code ecosystem with Google Cloud adds a second piece to that picture.
If Open Telco AI is about models and evaluation, Network as Code is about exposing programmable network functions that those models and agents can actually act on.

Put together, the signal is clear: telecom AI is moving towards a stack where shared models sit on top of standardised APIs, not siloed tools buried inside individual departments.

What this means for you: ask every vendor two blunt questions.

The most concrete operator move in this set came from MTN and Huawei.

According to Ecofin Agency on 12 March, the two companies expanded their partnership to modernise telecom networks, with the agreement covering AI-driven operations, broadband expansion, digital infrastructure, data monetisation, and ESG initiatives.

The deal’s central technical target is a shift towards Level 4 autonomous networks, using AI copilots and intelligent agents to move from human-led operations to closed-loop optimisation across planning, deployment, and operations.

The value of that story is not the slogan. It is the operating detail already attached to it.

MTN cited AI-enabled fuel savings at data centres in South Africa, dynamic energy management at cellular sites in Benin, automated fibre-cut detection in Côte d’Ivoire, and traffic balancing and optimisation in Nigeria. That portfolio spans energy, resilience, and traffic engineering rather than only customer-facing assistants.

Ecofin also noted that MTN had 301.3 million subscribers at the end of September 2025, which underlines the scale at which these operating changes could matter.

For HYPHN readers, the deeper point is that African operators are not waiting for perfect standards or perfect data estates before pushing AI into core operational tasks.

They are selecting problems with visible cost or uptime impact and turning those into autonomous-network stepping stones.

What this means for you: stop treating emerging markets as late adopters.
In this cycle, they may be the clearest proof points for AI tied to power efficiency, fault detection, and network stability.

A Reuters report on 2 March said Nokia was expanding partnerships with TIM Brasil and Deutsche Telekom as part of its AI technology push.

Separately, a 21 March statement said Circles and Huawei had signed a collaboration to advance AI-native digital telecom software.

Those two updates matter together because they show AI investment is no longer split neatly between “network AI” and “customer AI.”

The commercial case is becoming clearer as well.

A 12 March TM Forum article citing McKinsey research said operators are improving marketing conversions by up to 40 percent and reducing call-centre workloads by 25 percent in early deployments.

The same piece argued that dynamic offer agents can design bundles, test pricing scenarios, and launch offers within hours rather than weeks.

This is where many telecom strategies will fail if they stay too narrow.

If network teams build agentic automation while commercial teams still rely on fragmented pricing, care, and fulfilment stacks, operators will end up with two separate AI programmes and neither will deliver full value.

The smarter path is to connect monetisation, service assurance, and fulfilment around the same operating logic.

What this means for you: put your BSS roadmap in the same room as your autonomous-network roadmap.

The next serious gains will come when offer creation, service activation, assurance, and care share data and decision loops.