The nexus of Asia’s power demand and AI is getting stronger, with data centre investments set to double to US$200 billion ($254.22 billion) and tighten power markets a lot faster. These data centre investments will consume 0.5 trillion more power units by 2030. This could be as large as South Korea’s current power consumption, with significant implications for energy generation, distribution, and efficiency, not just for the power supply chain and global natural gas markets.
Power consumption continues to surprise our bullish expectations in Asia as AI adoption becomes more broad-based and tailwinds from onshoring and new-age manufacturing are sustained. We estimate Asia’s power consumption will grow at 5.0% CAGR, reflecting higher consumption in China, Japan, Malaysia, Thailand and India. Asia’s power consumption from data centres will rise to 100GW, with one-sixth of the incremental power demand in Asia to be consumed by data centres.
More efficient models to drive increased AI adoption across more use cases — such as advances in large language models (LLM) by DeepSeek-R1 — have demonstrated that cutting-edge AI capabilities can be achieved with significantly less hardware than other iterations and have defied conventional expectations of computing power requirements. We believe the commoditisation and commercialisation of foundation models influence the AI adoption curve. This accelerates global hyperscaler investments and the need for power.
In China, we see new data centre booking demand of 3GW–4GW per annum in the next three years, a 20% CAGR, with China’s AI graphics processing unit (GPU) self-sufficiency ratio rising to 82% in 2027, which will aid adoption and pace of rollout. In Southeast Asia, Malaysia remains a primary beneficiary of the buildout of data centres, with Tenaga’s pipeline growing to about 6GW (two-thirds of Southeast Asia’s data centre pipeline). We also see Thailand beginning to add capacity in a significant way, with over US$10 billion commitment by hyperscalers to develop AI and cloud infrastructure. India has 2GW of data centre capacity with plans to grow that to 5GW by 2030.
US power supply chain constraints will also accelerate Asia’s data centres and power demand. We estimate a 14GW potential shortfall in power for US data centres by 2028 to have significant spillover effects for Malaysia and the rest of Asia. With US power grids near capacity, we believe at least 6GW of training workload could be offshored to Asia by the decade’s end to unlock the “Time to Power” value from the US$1 trillion generative AI (GenAI) economy.
Spot power prices have reflected lower coal and gas prices, which have declined roughly 10% since the start of the year, but market tightness is reflected in power price over fuel spreads that have widened in multiple geographies, with consensus margins for Asian power generation companies rising 50 basis points (bps) year-to-date.
Policymakers have been calibrated in their approach to awarding more fossil-based power generation capacity and approving grid hardening investments, leading to only 7GW of new coal, gas or nuclear power generation capacity being announced in Asia ex-China last year. We believe that 10GW of restarts could be in the works as countries restart existing gas or nuclear power plants to keep power prices affordable.
See also: Alibaba expands AI cloud services in Malaysia, Philippines
Renewables alone cannot service data centre power demand due to intermittency. Natural gas is the solution that addresses the time to power as well as energy security. Gas-based power capacity can grow its utilisation rate and support the 24/7 power demand of data centres. It is interesting to note that Asia will drive consumption of more than 70% of new gas molecules produced globally as a shale revolution in the US gets exported to Asia. Gas share of generation will increase by 100 bps to 10% in Asia and should drive 25 million tonnes per annum of new gas consumption.
Gas can especially be combined with renewables (solar or wind generation in Asia) to reduce carbon intensity while providing quality power stability at competitive prices.
With Asia looking to account for two-thirds of the incremental growth in power consumption by 2030 and one-sixth of new power units consumed in Asia to be generated by gas, gas and renewable energy will have to work together to power the AI adoption curve.
Mayank Maheshwari is an Asean energy, materials & utilities analyst at Morgan Stanley
Read the main cover story:
