Singtel’s Next Act: Renting Out the AI Factory

The cloud has a surprisingly practical landlord. Singtel, the company generations of Singaporeans have paid to make calls and get online, now wants enterprises to rent the machinery behind artificial intelligence. Its offer reaches from network connections to server space, computing power and the people who make business software work. For a telecom group, that is a considerable extension of the monthly bill.

The change is already visible in concrete assets. Nxera, Singtel’s regional data-centre arm, announced the opening of its 58-megawatt DC Tuas facility in February 2026. In August, Singtel and KKR completed their acquisition of ST Telemedia Global Data Centres. Singtel emerged with an indirect 25% holding. The familiar phone company now has a larger interest in where other companies’ computing lives.

A telco with a second set of keys

Singtel still sells the ordinary things that make modern life function: mobile plans, home broadband, business connections and television services. Households buy access and entertainment; businesses buy connectivity and support. Optus supplies telecommunications in Australia. Regional investments extend the group’s economic reach through operators including Airtel and AIS, whose earnings contribute to Singtel’s results.

That structure explains the eye-catching customer numbers. Singtel describes a mobile footprint above 800 million across the group and its associates. Those are not all customers receiving Singtel-branded bills. An associate’s subscriber base represents market reach, while Singtel’s stake determines its economic interest. Confusing the two makes the company look simpler, and more directly enormous, than it is.

Its institutional history also resists a tidy founder myth. Singtel traces its roots to Singapore’s telephone services in 1879; the modern business was corporatised in 1992 and listed in 1993. Its own heritage account says impending telecom deregulation encouraged expansion overseas. The habit of looking beyond the home market is much older than the current enthusiasm for AI.

The business model mixes recurring subscriptions with enterprise network contracts, managed services, technology projects and infrastructure services. Regional associates add profit contributions from investments. At home, connectivity alternatives include StarHub, M1 and SIMBA. For corporate buyers, the comparison changes with the purchase: a mobile connection, an IT project and rented computing capacity each have their own competitors and contract economics.

The factory comes in four pieces

The newer offer is easier to understand as a set of purchases. Nxera supplies data-centre capacity. RE:AI, launched in October 2024, supplies AI computing and development resources. Paragon coordinates networks, edge computing and cloud services. NCS, the group’s technology-services business, helps government and enterprise clients build and operate systems. Each addresses a different headache on the journey from an idea to a working application.

RE:AI’s original pitch was access without having to buy and maintain an entire in-house setup. Its intended audience includes enterprises and public-sector organisations that need GPUs, storage, tools and network connections. For the buyer, the attraction is avoiding a miniature infrastructure business inside the organisation. The application, its data and the decisions about how to use it still need owners.

Paragon attacks a related problem: coordination. Introduced in 2022, it provides a common interface for deploying applications and managing network and computing resources. NCS supplies another kind of expertise, spanning applications, infrastructure, engineering, cloud, data and cybersecurity. In 2025 it announced a S$130 million AI transformation initiative across Asia Pacific. Hardware may be the photogenic investment; implementation is what turns it into something useful.

A security company puts the pieces to work

AETOS provides a concrete example. Singtel describes the security company’s Integrated Command Centre as using its 5G network, Paragon and multi-access edge computing to address operational silos. The point is to bring information together and process it where it can support operations. That is a more useful explanation of enterprise 5G than simply promising faster downloads.

The competitive argument is coordination across assets Singtel already operates or supplies. Its network infrastructure can connect an organisation’s sites, its computing services can support workloads, and its technology teams can help with implementation. A buyer could assemble comparable functions from separate providers. Singtel’s pitch is that making those pieces work together deserves a place in the purchasing decision.

The FY2026 Digital InfraCo review describes partnerships with NVIDIA on an applied-AI Centre of Excellence, with Cohesity on data resilience and with Mistral AI on models for sovereign environments. These are complementary capabilities, not proof of a universal advantage. Singtel’s strongest case is where customers need computing, local deployment and network performance to be considered together.

The expensive part is very real

AI infrastructure comes with bills well before it produces a customer result. Singtel’s FY2026 chief financial officer review sets out approximately S$3 billion of expected capital expenditure for FY2027, compared with S$2.5 billion the previous year. The growth spending is directed principally at Nxera and RE:AI. The review also says RE:AI investment is supported by secured customer contracts, limiting GPU inventory exposure.

Outside capital helps. In 2023, a KKR-managed fund committed up to S$1.1 billion for 20% of Singtel’s regional data-centre business. In February 2025, Nxera DCT secured a five-year S$643 million green loan from five banks to finance DC Tuas development. That loan is a funding amount, not a published final construction bill. The distinction is small in print and large in a spreadsheet.

The STT GDC transaction is a separate commitment. The consortium agreed to pay S$6.6 billion for the remaining 82% stake; its announced implied enterprise value was about S$13.8 billion, including leverage and committed-project capital expenditure. Those are different measures, and neither is Singtel’s solo cheque. After completion on 31 August 2026, KKR held 75% and Singtel 25% through the acquisition structure.

DC Tuas shows why the physical details matter. It incorporates a submarine cable landing station and supports direct-to-chip liquid cooling. Nxera said more than 90% of capacity was committed before launch. That is evidence of contracted demand, not a guarantee of eventual returns. Power, equipment, customer delivery and operating performance must still come together.

Before the GPUs, the costly detour

Singtel has tried expanding beyond conventional telecom before. It sold advertising-technology business Amobee in September 2022 and Trustwave in January 2024. At the 2024 annual meeting, management said those exits removed more than S$200 million in annual operating losses, measured as EBIT. The retreat matters because it supplies a less glamorous explanation for the current strategy: diversification also requires deciding what to stop owning.

The reset began in 2021, before RE:AI. Then FY2024 brought S$3.1 billion of non-cash impairment charges across assets including Optus, the Asia-Pacific cybersecurity business and NCS Australia. Singtel’s financial discussion linked Optus Enterprise’s fixed-network impairment to customer churn and price erosion. These charges recognised reduced asset values; they were not an equivalent cash payment made that year.

There is no need to invent a single boardroom epiphany. The public sequence is clear enough: review struggling businesses, exit selected holdings, recycle capital and concentrate investment. The current infrastructure push sits closer to Singtel’s networks and enterprise relationships. That makes the rationale easier to follow. It does not make the economics automatic.

A profit needs its small print

For FY2026, ended 31 March, Singtel reported operating revenue of S$14.26 billion and net profit of S$5.61 billion. Underlying net profit was S$2.77 billion. A net exceptional gain of S$2.84 billion, mainly from partial Airtel stake sales, explains why the headline profit deserves a second reading. Selling part of an investment creates a different kind of result from earning more on recurring services.

The quarter ended June 2026 offers another useful contrast. Digital InfraCo revenue rose 19%, while Singtel Singapore revenue fell 3.1% amid price competition. Group underlying net profit reached S$831 million, up 21%. These figures support the case for seeking growth beyond domestic connectivity. They do not show that the newer business has replaced the old one.

A network still has to answer the call

The infrastructure story also includes serious operational failures. Following Optus’s nationwide November 2023 outage, Australia’s communications regulator found that 2,145 people could not access emergency calling. Optus subsidiaries paid more than A$12 million in penalties in 2024. The regulator also identified 369 missed welfare checks on people who had attempted emergency calls.

“Triple Zero availability is the most fundamental service telcos must provide to the public.”

Nerida O’Loughlin · ACMA chair · November 2024

The issue did not end there. On 30 July 2026, the regulator commenced Federal Court proceedings against Optus Mobile over the September 2025 emergency-call outage, alleging failures on 1,005 occasions. Those are allegations in proceedings, distinct from the earlier penalties. For a group selling dependable infrastructure, reliability has to remain a central operating obligation alongside expansion.

Borrow the discipline, then price the workload

For readers considering Singtel, the useful lesson is to start with the job. A household may need a broadband connection and support. A business may need managed connectivity. A government agency considering AI may need computing, implementation and control over deployment. The breadth of the group’s catalogue becomes useful only when matched to a specific problem.

Our reading of the investment strategy suggests a practical procurement rule: establish demand before committing to expensive capacity. Ask for a workload-based quote that includes storage, networking, software, support and integration. Compare it with using existing infrastructure and with other providers. Test response time and recovery arrangements under the conditions the application will actually face.

A combined supplier can reduce coordination work, but it can also concentrate dependency. If an application needs little computing, already runs adequately elsewhere or gains nothing from local deployment, a larger infrastructure arrangement may add expense without enough benefit. Singtel can supply much of the machinery. The customer still has to decide whether the work going through it is worth doing.