Asia's AI Computing Expansion Hits Power Bottleneck
N.R. Finch
Morgan Stanley estimates Asia-Pacific data-center power demand will exceed 100 GW by 2030, growing at roughly 23% CAGR — but the region produces only about a third of the energy it consumes, making electricity supply the primary constraint on AI deployment.
How much data-center capacity is Asia building?
Morgan Stanley estimates total installed capacity across Malaysia, Singapore, Japan, India and China could reach roughly 100 GW.
Data-center power consumption is projected to rise from 193 TWh in 2023 to 832 TWh by 2030 — a ~23% CAGR, roughly matching the U.S. pace.
This means → by 2030, data centers will account for 3% of Asia's total power demand but 15% of all *incremental* demand — the pressure on new supply far outweighs the share of existing load.
Why is power scarcer than floor space?
A single AI GPU rack now draws 40–200 kW, versus just 5–15 kW for a traditional enterprise rack. Next-generation platforms will push toward several hundred kW per rack.
In plain terms = one AI cabinet consumes as much electricity as ten to dozens of conventional cabinets. The bottleneck has shifted from "finding a building" to "connecting to the grid."
This reflects a fundamental change: the binding constraint on data centers is no longer space but power density, cooling and grid access. Power-infrastructure readiness now dictates delivery timelines.
Can Asia's own energy supply keep up?
IEA data shows Asia's energy consumption grew 50% over the past decade, yet investment in conventional energy supply chains has fallen to historic lows — now below data-center investment in scale.
Asia consumes roughly half of the world's energy but produces only about one-third of what it needs domestically.
This means → the AI buildout is stacking on top of an already strained energy base. The supply-demand imbalance will only widen as compute demand accelerates.
Where will the money flow?
Morgan Stanley projects that energy security and AI power needs will together drive a $5 trillion+ investment supercycle across Asia, unlocking roughly $9 trillion in value creation.
Regional energy capex is expected to nearly double by 2030, spanning coal, natural gas, renewables, battery storage and the grid. Power generation (including coal) will account for more than two-thirds of the total.
Energy-storage demand will expand in tandem — annual incremental storage deployment tied to data centers could reach 321 GWh by 2030, equivalent to twice current storage demand.
What is the key variable for AI deployment pace?
Morgan Stanley's core thesis: the exponential growth of AI infrastructure is on a collision course with the physical world's energy constraints.
In plain terms = chips can iterate on Moore's Law timelines, but power plants and grids are built on multi-year cycles. The speed gap between those two curves is the bottleneck.
When that power bottleneck can be eased — through coordinated expansion of coal, gas and renewables — will be the critical checkpoint for Asia's AI compute rollout.
Content is for reference only, not financial advice.