Singtel building Southeast Asia’s AI power grid through Johor

• Singtel stacks data centres, GPU cloud, and a new Nvidia partnership for sovereign provision.
• CoE for Applied AI a software layer on a hardware bet for Singapore, Malaysia, and Indonesia.

Sovereign AI infrastructure doesn’t get built overnight, and Singtel isn’t pretending otherwise. The announcement on February 24 of a Centre of Excellence (CoE) for Applied AI with Nvidia – set to open at Punggol Digital District in roughly three months – is the latest piece in what Digital InfraCo CEO Bill Chang has been openly describing as an “AI grid”: a system modelled on how power utilities work, where large data centres act as generators, and fibre and 5G networks serve are the transmission lines delivering compute on demand.

It’s a useful analogy, and not rhetorically. The infrastructure being assembled to support it is considerable.

The grid, piece by piece

Nxera, Singtel’s regional data centre arm, opened its largest facility yet – DC Tuas in Singapore – on February 9, adding 58MW of AI-ready capacity and bringing the company’s total Singapore footprint to 120MW. More than 90% of DC Tuas was committed to customers before the doors opened.

Additional AI-ready capacity is scheduled to come online in Johor and Batam in the second half of 2026. By the mid-term, Nxera’s operational and pipeline capacity in the region is expected to exceed 400MW, up from 200MW today – with 40-50% of that configured specifically for high-density AI workloads, a share the company is actively pushing toward 60%.

Linked to Singapore via submarine cable, the Johor node forms part of what Chang describes as the generation backbone of the broader AI grid. It also positions Singtel to serve enterprises and government agencies in the Causeway looking for sovereign AI infrastructure without routing sensitive workloads through data centres they don’t control.

In early February, a KKR-led consortium with Singtel agreed to acquire the remaining 82% stake in ST Telemedia Global Data Centres for S$6.6 billion, a deal valuing the company at S$13.8 billion and described as the largest M&A transaction in Singapore for four years.

STT GDC operates 2.3GW of design capacity in 12 markets in Asia Pacific, the UK, and Europe. Once completed, expected to be in the first half of 2026, Singtel’s data centre exposure extends beyond Southeast Asia.

The CoE – after the press release

Against that backdrop, the Nvidia CoE looks like the applied intelligence layer sitting on top of a deliberate infrastructure build-out. The problem it’s designed to solve is that enterprises in Singapore and the broader region have been struggling to turn AI pilots into production systems.

Infrastructure complexity, skills gaps, and the pace of model development have all contributed to the bottleneck. The CoE will operate as a hands-on environment where organisations can bring actual problems, experiment using Nvidia’s platforms and specialists with Singtel’s RE:AI sovereign AI cloud, and build a path to deployment.

Chang has been candid about the gap: “We have seen many organisations struggle to move from the excitement of AI to meaningful impact.” Nvidia’s Ronnie Vasishta, SVP of Telecommunications, framed the partnership around the specific challenge of moving from trials to scaled deployment with confidence – language that reflects an industry-recognition that proof-of-concept success rates haven’t translated into production outcomes at scale.

The sovereign AI component of RE:AI also speaks to a regulatory and governance reality that is shaping enterprise AI decisions in the region. Banks and government agencies, in particular, face data residency obligations that make offshoring AI workloads complicated.

RE:AI is positioned to address that – and the demand signal has been strong. According to Singtel, the platform reached very high use shortly after launch in Singapore, driven primarily by government agencies and research institutions.

Preparing for the next wave

One of the four focus areas for the CoE is preparing Nxera’s data centres for the next generation of Nvidia GPUs. Nxera currently runs Nvidia GB200 systems at around 200kW per rack – already 20 to 25 times the density of a conventional data centre.

The next wave of Nvidia systems, codenamed Rubin Ultra and Feynman and expected between 2027 and 2029, could require between 600kW and 1MW per rack. That is the infrastructure challenge that makes sovereign AI infrastructure planning genuinely difficult: you are building for workload densities that don’t exist yet, in a market where power availability is already constrained.

The CoE is, among other things, where that future-proofing gets worked out in practice.

The Nvidia partnership and the CoE launch are scheduled for a June opening. Whether the grid metaphor holds – whether sovereign AI infrastructure in Southeast Asia ultimately flows like electricity, on demand, in borders – is still a question of execution. But the physical foundations are going in fast.

Want to experience the full spectrum of enterprise technology innovation? Join TechEx in Amsterdam, California, and London. Covering AI, Big Data, Cyber Security, IoT, Digital Transformation, Intelligent Automation, Edge Computing, and Data Centres, TechEx brings together global leaders to share real-world use cases and in-depth insights. Click here for more information.