Can liquid cooling solve Singapore’s AI infrastructure paradox?

• Nxera’s 58MW DC Tuas achieves 1.25 PUE using Singapore’s largest liquid cooling data centre deployment, proving tropical climates can support AI workloads efficiently
• Over 90% pre-committed capacity signals desperate demand in a market where vacancy rates sit below 2% and new approvals require 50% green energy sourcing

When Singapore imposed its data centre moratorium in 2019, the message was clear: the city-state’s 735 square kilometres couldn’t sustain endless rows of power-hungry facilities consuming 7% of national electricity.

Six years later, Nxera has opened DC Tuas with a solution that challenges the assumption that AI infrastructure and sustainability are mutually exclusive—by deploying Singapore’s largest liquid cooling data centre infrastructure at commercial scale.

The 58MW facility, which brings Nxera’s total Singapore capacity to 120MW, was over 90% committed before it even opened. That’s not surprising in a market where vacancy rates hover below 2%.

What’s remarkable is how DC Tuas achieves a power usage effectiveness (PUE) of 1.25 in a tropical climate—significantly better than the 1.4-1.6 range that most regional facilities struggle to break.

The difference lies almost entirely in its cooling architecture. DC Tuas hosts what Nxera describes as Singapore’s largest direct-to-chip liquid cooling deployment in a multi-tenanted facility. This technology has rapidly evolved from experimental to essential as AI workloads have exploded in density and power consumption.

Why can’t air cooling keep up with AI?

The physics are straightforward: NVIDIA’s H100 GPUs consume around 700W per chip, while the newer Blackwell architecture pushes past 1,000W. Multiply that across entire racks running at 40kW to 120kW, and traditional air cooling simply runs out of thermal capacity. Air would need to be either sub-zero in temperature or moving at gale-force speeds to extract that much heat—neither practical nor efficient.

Water’s thermal conductivity is approximately 23 times higher than air’s. Direct-to-chip liquid cooling leverages this by circulating coolant through cold plates mounted directly onto CPUs and GPUs, absorbing heat at the source rather than trying to move it through metres of hot air. Research from Florida Atlantic University, recently highlighted by NVIDIA, demonstrates that liquid-cooled GPU systems deliver up to 17% higher computational throughput while reducing node-level power consumption by 16% compared to air-cooled equivalents.

For DC Tuas, this translates to practical advantages: higher rack densities without thermal throttling, reduced reliance on energy-intensive chillers, and the ability to operate with warmer facility water temperatures—critical in Singapore’s tropical environment where ambient temperatures make conventional cooling exponentially more expensive.

Engineering for constraints

Bill Chang, CEO of Nxera and Singtel’s Digital InfraCo unit, frames DC Tuas as a response to Singapore’s unique constraints. “The ability to deploy higher-density, compute-intensive AI workloads sustainably is increasingly critical for a strategic market where data centre capacity is constrained,” he noted at the launch.

Beyond liquid cooling, DC Tuas incorporates multiple systems designed specifically for tropical efficiency: smart thermal features, high-efficiency electrical systems, solar power generation, rainwater harvesting, condensate reuse, and blowdown recovery.

The facility holds Green Mark Platinum certification from Singapore’s Building and Construction Authority and IMDA, meeting the stringent sustainability requirements that now govern all new data centre capacity in Singapore.

The 120,000-square-foot, eight-storey facility also distinguishes itself as Singapore’s only hyperconnected data centre integrated with a cable landing station, providing customers direct access to international and domestic networks—essential for latency-sensitive AI training and inference workloads.

The regional calculation

Singapore’s December 2025 call for applications (DC-CFA2) makes the trade-offs explicit: operators seeking 200MW of new capacity must demonstrate at least 50% green energy sourcing, best-in-class PUE ratings, and “significant contributions” to Singapore’s economic objectives.

The March 2026 application deadline will reveal which operators believe they can meet these requirements profitably. Nxera is already looking beyond Singapore’s borders. Additional AI-ready capacity is scheduled for Batam and Johor in the second half of 2026, part of a strategy to more than double operational and pipeline capacity from 200MW in 2026 to over 400MW mid-term.

This regional approach acknowledges that while Singapore’s regulatory framework creates high barriers to entry, it also creates pricing power and competitive moats for operators who can meet the technical requirements.

The broader question is whether other land-constrained markets facing similar resource pressures can replicate Singapore’s model: use scarcity strategically, impose technical requirements that filter for quality, and trust that demand will adjust rather than disappear. DC Tuas suggests the answer depends heavily on operators’ willingness to deploy advanced cooling technologies at a commercial scale—and accept the capital costs that come with them.

For enterprises and hyperscalers betting on AI while facing pressure to reduce environmental footprints, facilities like DC Tuas demonstrate that the infrastructure exists to support both goals simultaneously. The challenge is finding operators willing to build it.

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.

TechHQ is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.