At what rack density does liquid cooling become necessary?

Uptime Institute's global data center survey found that direct liquid cooling has moved from a niche option to a mainstream consideration once rack density exceeds around 20 kW, a threshold Southeast Asia's warmer, more humid climate tends to reach sooner than temperate markets.

Why is cooling cost per megawatt rising alongside capacity?

Average cooling cost per MW is projected to climb from 1.15 in 2024 to 1.70 by 2033 as operators replace lower-cost air-based systems with more expensive liquid and hybrid infrastructure to handle AI-driven rack densities.

What is the biggest constraint on AI-driven data center growth in the region?

Grid access and transmission capacity, not cooling technology procurement, are the primary bottleneck. Malaysian utility TNB is raising capital expenditure to roughly RM42.8 billion between mid-2025 and 2027 specifically to expand transmission for data center demand in hubs like Johor.

Published: July 22, 2026

How AI Workloads Are Accelerating Growth in the Southeast Asia Data Center Cooling Market

AI Is Rewriting the Thermal Math of Southeast Asia's Data Centers

Southeast Asia's data center industry built its reputation on connectivity and cost efficiency. That reputation is now being rewritten by heat. AI training and inference workloads pack far more silicon into the same rack footprint than the cloud and enterprise workloads the region's facilities were originally engineered around, and every one of those extra chips has to be cooled. According to Epignosis Insights, installed cooling capacity across Southeast Asia stood at 1,800 MW in 2024 and is projected to reach 3,763 MW by 2033, expanding at a compound annual growth rate of 9.2 percent between 2027 and 2033. The growth rate itself is accelerating year over year, climbing from roughly 6.1 percent in 2025 to more than 10 percent by 2033, a pattern that tracks closely with the region's AI infrastructure buildout rather than general-purpose cloud demand.

From Kilowatts to Megawatts: Why Rack Density Changes Everything

The pressure is structural, not cyclical. A conventional enterprise rack draws a fraction of the power a modern AI accelerator rack requires, and that gap keeps widening. Uptime Institute's global data center survey found that average rack power density climbed sharply as AI clusters entered mainstream deployment, with a growing share of operators reporting racks that air cooling alone can no longer service. The same research found that direct liquid cooling had already moved from a niche deployment to a mainstream consideration among operators tracking density above 20 kW per rack. In Southeast Asia specifically, where ambient temperatures and humidity are already higher than in temperate markets, that threshold arrives sooner and pushes operators toward liquid and hybrid cooling architectures earlier in a facility's design life than in Europe or North America.

The Cost Curve Is Climbing Alongside Capacity

Cooling AI-dense infrastructure is not just a technical shift, it is a financial one. Epignosis Insights data shows the average cooling cost per MW rising steadily from 1.15 in 2024 to a projected 1.70 by 2033, a trajectory that mirrors the broader market's expansion from an estimated USD 2,070 million in 2024 to USD 6,397 million by 2033, a 12.8 percent compound annual growth rate between 2027 and 2033. That combination, more installed capacity and a higher cost per megawatt, signals that operators are not simply scaling existing designs. They are replacing lower-cost air-based systems with materially more expensive liquid and hybrid infrastructure to keep pace with rack densities that traditional cooling was never engineered to handle.

Investment and Grid Reality Are Colliding

Government data underscores how fast capital is moving into the region even as physical infrastructure struggles to keep up. Malaysia's Investment Development Authority reported that data centers and cloud computing accounted for RM34.6 billion across 33 approved projects in the first quarter of 2026 alone, part of a cumulative RM144.4 billion approved between 2021 and mid-2025. Consulting firm Wood Mackenzie found that data center demand in Johor, Malaysia's leading hub, has already reached approximately 3.8 GW, representing over half of total data center demand across Peninsular Malaysia, while flagging substation and transmission bottlenecks as the primary constraint on new capacity. Trade press coverage from DatacenterDynamics reported that national utility TNB is raising its capital expenditure allowance to roughly RM42.8 billion between mid-2025 and the end of 2027 specifically to expand transmission and distribution capacity for this demand. The result is a market where cooling technology decisions are increasingly shaped as much by grid access as by thermal engineering.

Vendors Are Positioning for a Structural, Not Temporary, Shift

Equipment suppliers are treating this as a durable change in demand rather than a short-term surge. Vertiv's own annual report disclosed that its combined order backlog reached USD 15.0 billion at the end of 2025, more than double the USD 7.2 billion recorded a year earlier, a jump the company attributed directly to AI-driven infrastructure orders outpacing its ability to fulfill them. That kind of backlog growth among a major thermal management supplier reflects what is happening on the ground across Southeast Asia's newest campuses: operators are locking in liquid cooling and high-density thermal systems years ahead of commissioning, not retrofitting them after the fact.

What This Means for the Region's Data Center Strategy

Southeast Asia's data center cooling market is no longer growing simply because more facilities are being built. It is growing because the facilities being built now run hotter, denser, and more power-intensive workloads than anything the region has hosted before. Markets like Johor, Batam, and parts of Thailand are absorbing this demand fastest, but each faces a similar constraint: cooling technology can be procured and installed far more quickly than transmission infrastructure can be expanded. For operators, investors, and policymakers, the AI-driven cooling boom is as much a grid and capital planning challenge as it is an engineering one, and the markets that align power availability with cooling ambition, not just those with the lowest land and construction costs, will lead this next phase.

Frequently Asked Questions

Why are AI workloads driving cooling demand more than traditional cloud workloads?

AI accelerator racks concentrate far more compute and power draw per square meter than conventional servers, pushing rack densities past the point where standard air cooling remains effective.

Is liquid cooling now standard across Southeast Asia's data centers?

Not yet standard across the board, but adoption is accelerating quickly in new AI-focused builds, particularly where rack density exceeds roughly 20 kW, a threshold industry research increasingly treats as the point where liquid cooling becomes necessary rather than optional.

Which Southeast Asian markets face the tightest cooling and power constraints?

Johor in Malaysia is the most cited example, where data center demand already represents over half of the state's total electricity demand and substation capacity has become the binding constraint on new projects.

Are cooling costs expected to keep rising through the rest of the decade?

Yes. Projections show average cooling cost per megawatt rising steadily through 2033 as the market shifts from lower-cost air cooling toward liquid and hybrid systems required for AI-density racks.

How are governments responding to the strain on transmission infrastructure?

Utilities and regulators across the region are raising capital expenditure for grid upgrades, tightening approval processes for new facilities, and in some cases pausing approvals for non-AI or high-water-use projects while transmission capacity catches up with demand.