AI lifts Asia Pacific data centre costs, report says

Asia Pacific data centre construction costs are diverging as artificial intelligence changes how facilities are built. Cushman & Wakefield's latest cost guide puts the regional range at USD $7.9 million to USD $19.2 million per megawatt.

The report found a 2.4-fold gap between the cheapest and most expensive markets. Japan ranked highest at USD $19.2 million per MW, followed by Singapore at USD $17.9 million, while Taiwan was the lowest-cost market at USD $7.9 million per MW.

The findings point to a sharper market-by-market split in development economics as operators and investors plan larger campuses for AI workloads. Differences in sourcing strategies, labour costs and supply-chain conditions are widening the gap across the region.

Japan and Singapore also recorded the strongest annual increases among the largest hubs. Year-on-year costs rose 20.9% in Japan and 22.7% in Singapore, compared with 11.5% in South Korea and 3.8% in Australia.

Elsewhere, Thailand posted a 14.7% annual increase, the Philippines 17.7%, while the Chinese Mainland rose 0.4%, New Zealand was flat, and Taiwan increased 1.7%.

AI designs

The shift is tied to changes in data centre design as AI systems demand more electricity, different cooling systems and stronger structures. These requirements are directly affecting shell-and-core construction costs, especially in markets with constrained power access and limited labour availability.

"Across Asia Pacific, construction cost inflation diverges sharply, with some markets seeing increases above 15% while others remain below 5%," said Andrew Green, Head of Data Centre Group, Asia Pacific, Cushman & Wakefield.

"A key reason for this split is that AI is reshaping the physical and technical requirements of data centres, particularly at the shell and core level. Higher power density, more complex cooling systems and stronger structural requirements are becoming standard in AI-ready facilities, with very different cost implications depending on local power availability, labour capacity and delivery conditions," Green said.

Facilities are increasingly being planned around higher-density computing and advanced cooling methods, creating a new baseline for new developments. That is forcing developers to reassess both site selection and project budgets as demand for AI infrastructure rises.

"AI is transforming data centre design far faster than traditional development cycles anticipated. Each new generation of high-performance hardware demands more power, more cooling and greater structural resilience. These requirements are redefining what it means to build a future-ready facility, and markets that can meet them effectively are pulling ahead while others face rising delivery and cost pressures," said Pritesh Swamy, Head of Research and Advisory for the Data Centre Group in Asia Pacific, Cushman & Wakefield.

Power constraints

In established hubs such as Tokyo, Singapore, Sydney, Taipei, and Johor, competition for sites with access to power is increasing development pressure. Grid constraints and longer connection timelines are increasing both complexity and execution risk.

That matters because power availability has become central to project viability, particularly for AI-focused developments that require higher-density infrastructure. In several of the region's key markets, land is no longer the main limiting factor.

Procurement is also becoming more difficult. Price gaps between Chinese and non-Chinese suppliers are widening, while longer equipment lead times and broader use of prefabricated and modular methods are adding more variability to budgets.

Some existing facilities are also proving hard to adapt. Older data centres built for earlier workloads may not be suitable for dense AI computing, pushing owners to consider alternative roles such as edge computing, warm storage or interconnection hubs.

"Developers across the region are navigating a more complex delivery environment. Power readiness, procurement decisions and site conditions have become central to project feasibility, particularly for AI-focused builds that require higher-density infrastructure and faster deployment timelines," said Sam Asher, Head of Development and Commercial Advisory, Project and Development Services, Asia Pacific, Cushman & Wakefield.

Regional rankings

By mid-range build cost, Japan ranked first at USD $16 million per MW, followed by Singapore at USD $14.4 million and South Korea at USD $10.6 million. Australia stood at USD $10 million, Hong Kong at USD $9.8 million and Malaysia at USD $9.6 million.

New Zealand came in at USD $9.3 million, ahead of Thailand at USD $8.8 million, Indonesia at USD $8.3 million, and the Philippines at USD $8.2 million. India was ranked at USD $7.4 million, Vietnam at USD $7.2 million, the Chinese Mainland at USD $7.1 million and Taiwan at USD $6.5 million.

The report suggests operators pursuing large AI deployments will need increasingly detailed local modelling rather than broad regional assumptions. In Asia Pacific's data centre market, access to power, supply chain resilience, and the cost of building for AI are now determining where new projects can proceed.