← Insights
Balance ReportSector study · 4 chapters · 1 chart·For CMO, CDO, Chief Customer Officer, CIO

The Sovereign AI Balance Report: Telco Edition

Churn and ARPU are decided by how well a telco reads its subscriber data, and that data is the moat. Where telco customer AI belongs on the sovereignty spectrum under NIS2, and why feeding the graph to a frontier API rents out the moat. With a modelled portfolio benchmark.

Of telcos call AI the new normal; subscriber data is the gap (McKinsey)
16%
Telco AI value that belongs sovereign (modelled)
~60%
Subscriber data sent off-perimeter to save on compute
0
Chapter 01

The subscriber graph is the moat

A telco grows by cutting churn and lifting ARPU, and both run on its view of who connects to whom, where, and how. After the Reset, the temptation to feed that to a frontier API is the temptation to rent out the moat.

Telcos hold a uniquely rich and sensitive dataset, location, communication patterns, device and network behaviour across an entire subscriber base. It is what decides the numbers the business lives on, which offer keeps a subscriber from leaving, which upgrade lifts a household's ARPU, which journey converts, and it is exactly the data that must not leave the perimeter to power AI. The Reset makes sovereign inference over that graph viable; the growth case makes it necessary.

Telcos are also critical infrastructure under NIS2, with resilience and supply-chain obligations that a third-country dependency complicates. The sector's instinct for operational control is, for once, perfectly aligned with the sovereign-AI thesis.


Chapter 02

Where telco AI belongs

Subscriber-data workloads home; network ops mostly sovereign; a narrow frontier edge.

At the sovereign end: churn prediction and retention offers, personalisation and next-best-offer over subscriber data, fraud modelling, network-operations AI, and security analytics, high-volume, high-sensitivity, and the direct levers of growth, best run on self-hosted models inside the perimeter. At the frontier edge: customer-service assistance where conversations can be kept non-identifying, marketing content, and public-data analysis. The split lands around sixty per cent sovereign.

The volume economics reinforce the sovereignty case: at subscriber-base scale, self-hosted inference is not only safer but, fully costed, cheaper than per-token frontier pricing. The moat and the margin point the same way.

Figure · Recommended AI portfolio split, representative European telco
Self-hosted open-weight
48
Frontier in EU / private
20
Hybrid routing
18
Frontier API
14

Share of AI workloads by placement, weighted by subscriber-data sensitivity and critical-infrastructure duties. A Rindogatan-modelled benchmark, not survey data.


Chapter 03

Why the data must stay home

Feeding the subscriber graph to a frontier API trades a durable advantage for a convenience.

The subscriber graph is valuable precisely because no competitor has it. Sending it outside the perimeter to power AI, to a model and a jurisdiction the telco does not control, erodes the one asset that differentiates the business, and does so to save on infrastructure that, at this scale, is cheaper to run sovereignly anyway. It is the clearest moat-for-bridge trade in any vertical.

Sovereign inference keeps the graph where it belongs and still delivers the AI capability, personalisation, fraud detection, network optimisation, that the data enables. Claude has a place at the non-personal edge; the subscriber data has only one home.

A telco's subscriber data is its most defensible asset. Sending it to a frontier API to save on infrastructure is selling the moat to rent the bridge.


Chapter 04

The telco's 90 days

Four moves to AI that strengthens the moat instead of leaking it.

Classify subscriber-data workloads as sovereign-by-default and identify the high-volume cases, churn and retention offers, fraud, network ops, where self-hosting wins on cost and control at once. Stand up sovereign inference behind the existing platform, and prove the customer case first: a point of churn is visible on the P&L within a quarter.

Keep the frontier model to the non-identifying edge, prove one sovereign workload at subscriber scale, and govern placement so no team routes the subscriber graph off-perimeter. The moat is the strategy.


Sources & methodology
  • 1. Headline figures are Rindogatan models, directional benchmarks to be calibrated to a specific institution, not survey statistics.
  • 2. Partner data points are drawn from publicly published research (e.g. Snowflake's Modern Marketing Data Stack, Databricks' State of Data + AI) and cited for direction only.
  • 3. Regulatory references: EU AI Act, Reg. (EU) 2024/1689; GDPR, Reg. (EU) 2016/679; DORA, Reg. (EU) 2022/2554; NIS2, Dir. (EU) 2022/2555.
  • 4. Sovereign deployment modelled on European sovereign infrastructure.