Singtel bets on sovereign AI premium telecom services

James Calloway

Singtel operates maybe five, 10 or 15 data centers, compared with the hundreds or thousands run by hyperscalers, a scale advantage it says can support premium sovereign AI services.

"It's easier for providers like us to build best-in-breend GPUs and infrastructure because we are operating maybe five, 10 or 15 data centers, compared with hyperscalers that operate hundreds or thousands. We can test infrastructure quickly and also offer stringent Service Level Agreements (SLAs) and data-locality guarantees."

— Vinod Joseph, Group Vice President, AI and Data Strategy at Singtel

Singapore's largest service provider is testing where sovereign AI can extend telecom revenue beyond raw GPU infrastructure. The opportunity, discussed by Joseph at FutureNet Asia, covers AI models, intelligent routing and systems that preserve user context. He said customers will need choice because geopolitics makes dependence on one model or tool set too restrictive.

Joseph also said Singtel must continuously assess model performance against each use case and expected value. A carrier cannot treat access to frontier models, open-source models and vendor tools as interchangeable commercial products. Control over selection matters as much as access.

That makes sovereign AI more than a data-residency label. Joseph defined sovereignty as control of assets, with data localization one element and flexibility across tools, vendors and solutions another. Several countries have imposed data-localization regulations, while geopolitical tension is pushing enterprises and organizations toward sovereign infrastructure.

What does this change for telecom operators?

It moves the commercial proposition from selling hosted compute to managing a controlled set of AI services. Singtel is considering four linked capabilities:

• Process sensitive data locally in Singapore while allowing other workloads to run outside the country.
• Combine owned infrastructure with hyperscaler and partner capacity through a "hybrid mesh architecture."
• Place control points over how customers use data and models.
• Redirect workloads between frontier and open-source models to meet data-locality requirements.

Joseph argues that control over data, models and infrastructure can differentiate Singtel in an increasingly competitive market. The operator also sees value in testing infrastructure quickly and offering stringent service levels and data-locality guarantees, rather than competing only on raw GPU availability.

Control does not remove the compliance burden. Singtel must provide customers with meaningful authority over where workloads run and which models process them while satisfying the rules applying to each jurisdiction. Its proposed hybrid design attempts to reconcile those requirements without restricting every workload to infrastructure inside Singapore.

International operators face a related problem. Andrew Bennett, CTO for AMEA at BT, used the same FutureNet Asia panel to question whether one sovereign-AI product can serve multiple countries.

"Sovereignty and compliance are largely defined at the nation level and there is no common model that works across all countries yet. For international telecom groups, the challenge is to scale solutions developed for individual markets into products and offerings that can serve customers globally."

— Andrew Bennett, CTO for AMEA at BT

BT's warning sets a limit on standardization. A service designed around Singapore's controls may not transfer directly to another market, so multinational carriers would need common control frameworks while retaining country-specific infrastructure, vendor arrangements and compliance processes.

The commercial proof remains limited. The material identifies no launched sovereign-AI package, customer commitment, price, revenue target, capital commitment or rollout date. Singtel's stated scale of five, 10 or 15 data centers provides an operating base, while its claims of best-in-breed infrastructure and premium differentiation remain propositions for the market to test.

A separate Verizon deployment. Verizon separately unveiled a 1-Gig fixed-wireless access service using millimeter-wave technology, intended to add 10 to 15 million high-speed broadband premises. Verizon called the technology a "breakthrough" for matching fiber speeds and reported that overall third-quarter subscriber growth surged, without providing a subscriber figure.

For now, rollout timing, infrastructure spending and customer uptake remain undisclosed. Those details will determine whether sovereign AI becomes a billable premium service for Singtel and a scalable product for multinational carriers, or remains a strategic option.