-01.jpg)
Behind The Scenes: Microsoft: Data Centres More Efficient Than Public Perception Suggests
Data centres have become one of the most debated pieces of infrastructure in the modern economy. As artificial intelligence (AI) drives a wave of hyperscale investments across Southeast Asia, public concern about water consumption, energy demand and grid pressure has intensified.
Hyperscalers such as Microsoft argue that modern facilities are far more efficient than public perception suggests, and that the industry’s relationship with national grids is more symbiotic than extractive.
Data centre capacity in Southeast Asia is projected to triple from 2025 levels by 2030, driven by a tenfold surge in AI use, according to a KPMG report.
For Microsoft, which operates more than 400 data centres across 70 regions globally, the conversation starts with visibility, or the lack of it.
Alistair Speirs, general manager for Azure Infrastructure at Microsoft, says much of the public concern stems from infrastructure that most people never see.
“During the smartphone revolution, people saw devices evolve every year in their hands. Data centres happen in secure facilities, out of sight. Even IT professionals used to server rooms don’t see how dramatically infrastructure has changed,” he said when met during a media tour of Microsoft’s newly built data centre in Jakarta, Indonesia, in February.
That invisibility, Speirs says, obscures how much the technology has evolved. Microsoft applies a versioning system to its fleet with each new facility built on lessons from the ones before it, adapted for greater efficiency in power and water use, or designed around the next generation of compute innovations. A versioning system means each new data centre is built better, with improvements from the previous data centre designs.
The result, Speirs argues, is infrastructure that bears little resemblance to the server rooms most people picture when they hear the words data centre.
The concern over power use is not unfounded, however, as AI workloads are genuinely more power-intensive than traditional compute.
A Microsoft spokesperson says rising AI adoption is the primary driver.
“AI data centres consume more per unit because AI demand itself is rising rapidly. Adoption is accelerating. GPU workloads and model sizes are growing. So total consumption rises because demand rises,” the spokesperson says. The key metric, according to the spokesperson, is not total consumption but performance per watt — transactions per second per watt — which improves with each new generation of hardware.
In Malaysia, the grid reality compounds this concern as it still relies largely on fossil fuels, which sit at the heart of public concern that data centres, however efficiently designed, are drawing power from carbon-intensive infrastructure in markets where the renewable transition remains incomplete.
With power becoming a flashpoint, Microsoft’s response is its power purchase agreement model. Rather than the “behind-the-metre” approach favoured by some industry players, which involves contracting directly with power plants and bypassing the grid, Microsoft commits to long-term agreements with utility companies instead.
“If we commit to, say, 200MW over 10 years, utilities can finance new generation capacity. Because we only buy zero-carbon energy, that investment goes into solar, wind, hydro or geothermal,” Speirs says, adding that this model benefits the wider grid rather than simply securing cheap power for Microsoft’s own operations.
As data centres are also software-defined, Speirs says they can interact with the grid more flexibly than most industrial consumers.
“With backup systems, we can run on backup when the grid is stressed, or in some cases, send power back to support it,” Speirs explains. This ability to absorb and smooth supply fluctuations is particularly relevant as renewables, variable by nature, make up a growing share of regional energy mixes. “Rather than simply consuming energy, data centres can help create larger, more stable grids,” he adds.
Backup systems also present their own challenges. The uninterruptible power supply batteries and generators required to sustain operations during grid disruptions — which Speirs says can last up to a week in cases of major national disruption — still run primarily on fossil fuel.
In Indonesia, Microsoft’s generators run on biodiesel at about 40% non-fossil fuel content, meaning the remaining 60% is still fossil-derived. Longer term, Speirs points to hydrogen fuel cells as the most credible path forward as the company has already tested powering entire data centres for 48 hours using the technology. “The science is known, the challenge now is scaling,” he says.
On cooling design, Speirs addresses one of the most persistent public assumptions that data centres must be kept as cold as traditional server rooms.
“It’s not the temperature but the airflow that matters. The cold aisle isn’t especially cold. It’s close to outside ambient temperature. As long as air moves across the servers, they remain cool,” he says.
Water consumption is the other flashpoint, and one where the gap between perception and reality is sharpest. The common assumption is that liquid cooling uses more water than air cooling but Speirs explains that the opposite is true.
Air cooling relies on evaporation to dissipate heat, and that evaporation consumes water continuously. Liquid cooling, by contrast, operates as a closed loop with no evaporation. The same water circulates through the system for up to seven years without needing to be replaced.
As high-powered components such as GPUs, and eventually CPUs, shift to liquid cooling infrastructure, Speirs says overall water consumption across data centres will fall relative to the air-cooled facilities they replace. Rainwater harvesting is also used on site.
The scale of that consumption is becoming easier to quantify. According to a blog post by OpenAI chief executive Sam Altman, a single ChatGPT prompt uses about 0.34 watt-hours of energy and 0.322ml of water.
A Gemini prompt consumes slightly less — about 0.24 watt-hours and 0.26ml of water — according to a recent Google publication. Individually the figures appear modest, but multiplied across billions of daily queries, they add up quickly.
The International Energy Agency’s April 2025 report forecasts global data centre electricity consumption exceeding 945 terawatt-hours by 2030, which is a sharp rise.
Microsoft has committed to being water positive and carbon negative by 2030. For its data centres specifically, the company targeted net zero by 2025. Microsoft says the cycle has been completed and the verification report is now being finalised.
Microsoft’s Indonesia Central cloud region in Karawang, launched in May 2025 as the company’s first in the country and backed by a US$1.7 billion investment commitment, shares the same three-availability-zone architecture as the Malaysia West cloud region in Greater Kuala Lumpur, which went live the same month.
For Malaysia, specifically, the opening of Malaysia West introduces dimensions that go beyond efficiency metrics. Local data residency reduces latency and enables mission-critical workloads that government agencies and financial institutions were previously reluctant to place on cloud infrastructure hosted abroad.
“Malaysia West provides local data residency and low latency, which increases comfort for sensitive workloads,” a Microsoft spokesperson says.
On data ownership, the company is clear that customer data belongs to the customer. “We do not use enterprise customer data to train our models,” the spokesperson says.
Early anchor customers include Petroliam Nasional Bhd (PETRONAS), TNG Digital and Sirim Bhd, among others. A second Malaysian region in Johor is in development, planned to incorporate zero-water evaporation cooling and backup generators running on renewable biofuel.
Microsoft’s investment in Malaysia, US$2.2 billion committed in 2024 over four years, is the largest in its then 32-year history in the country.
It is projected to generate US$10.9 billion in new revenues and more than 37,500 jobs by 2028, including 5,700 skilled IT roles.
On skills, Microsoft says its talent development programmes in Malaysia have reached over one million people, though the company’s AIForMYFuture initiative carries a stated target of training 800,000 Malaysians by end-2025. “Infrastructure and workforce must advance together,” a spokesperson says. For small and medium enterprises, Microsoft says the opportunity extends beyond text-based tools into manufacturing optimisation, supply chains, production cycles and robotics.
Source: theedgemalaysia.com
