Asia Pacific Data Construction Cost Guide 2025

Vivek Dahiya Head of Advisory & Transaction Services, Data Centre Group, Asia Pacific

INTERNATIONAL MONETARY FUND ASIA PACIFIC AI READINESS COMPARISON

0.9

0.25

It is likely to be several more years before the full impact of AI is felt in Asia Pacific. The region trails both the United States and Europe in terms of AI deployment, although several operators have acquired land parcels and are actively planning their AI data centres. Regardless, lessons from other regions provide some early insights. Location Theoretically, learning AI data centres are location agnostic. But the challenges of establishing new infrastructure in remote locations means they are likely to continue being built in or around existing infrastructure and talent. To avoid cannibalizing the power of existing availability zones, proximate industrial sites could be suitable locations for learning AI data centres. Resource Availability The International Monetary Fund (IMF) has ranked major Asia Pacific markets including Singapore, Hong Kong, South Korea, Japan and Australia among the top 20 markets in both the AI Preparedness Index and the Digital

Infrastructure Index. However, challenges with land and/or power availability, especially in key data centre clusters, could mean operators look to alternative, emerging locations that have the power and water availability to support their AI deployments and long-term scalability. Data Sovereignty The introduction of data sovereignty laws could be one of the greatest challenges to AI deployment. It is likely that such laws will make it challenging to locate learning data centres in a single market. Operators may also need to comply with data reporting and sharing regulations introduced by respective governments. These laws are likely to significantly impact data centre location strategy. Design & Construction In response to the rapid uptake expected from AI, a number of data centres that were in either the planning or early development stages have since been modified into so-called ‘hybrid’ data centres capable of accommodating both cloud and AI requirements. These hybrid designs may

include greater loading on alternate floors and/or increased ceiling heights to house new cooling requirements. These accommodations have been made somewhat in advance of the cooling technologies themselves. While several forms of liquid cooling, including immersion and direct to-chip, are currently available, these technologies are yet to achieve full confidence among data centre operators and users, with research into their durability, maintenance, leakage and compatibility continuing. Obsolescence Despite the rapidly evolving landscape, the good news is that data centres built post-2020 remain largely relevant today. Continued demand from cloud firms and enterprise clients minimise the risk of obsolescence—a real consideration for investors given the significant amounts of capital involved in the sector. As technology continues to evolve, current best practice is to build for agility—the ability to consume and process the maximum number of megawatts and to allow end-users to deploy the hardware to support AI workloads.

0.8

0.2

0.7

0.6

0.15

0.5

0.4

0.1

0.3

0.2

0.05

0.1

0

0

India

Japan

Vietnam

Malaysia

Thailand

Australia

Indonesia

Singapore

Philippines

South Korea

New Zealand

Taiwan, China

China Mainland

Hong Kong, China

AI Preparedness Index

Digital Infrastructure Index

Source:

CUSHMAN & WAKEFIELD | ASIA PACIFIC - DATA CENTRE CONSTRUCTION COST GUIDE 2025

7