AI data centres reshape infrastructure costs as market splits

The global data centre market is undergoing significant structural changes, with artificial intelligence (AI) driving a divergence in construction costs and operational models, according to Turner & Townsend’s latest Data Centre Construction Cost Index.

Global data centre revenues are projected to reach US$527.5 billion this year, fuelled largely by accelerated investment in AI. While optimism remains high, 73% of industry respondents describe the sector as ‘recession-proof’, the report warns that growing challenges around planning, delivery, and operations could impact future growth.

Turner & Townsend’s survey, which draws on insights from 52 global markets and a broad cross-section of industry professionals, highlights the growing complexity of data centre developments and the critical need for investors to understand cost dynamics across regions and technology types.

Construction costs for traditional cloud-based data centres have begun to stabilise, with a 5.5% year-on-year increase in cost per watt, down from 9.0% in 2024. This moderation is attributed to a general easing in construction cost inflation and maturing local supply chains in several regions.

However, data centres built to support AI workloads present a different picture. These facilities, typically featuring high-density, liquid-cooled environments, carry a premium of 7% to 10% over similarly sized, air-cooled facilities in the US, according to Turner & Townsend’s analysis using its Hive platform.

The AI premium reflects the more complex technical and cooling systems required to support intensive AI processing. These builds are gaining traction in the US, UK, Europe, and East Asia, where demand for large-scale AI infrastructure continues to rise.

Despite their higher construction costs, AI data centres benefit from operational flexibilities not typically available to traditional cloud environments. Cloud data centres are built with stringent redundancy measures, multiple backup power sources and extensive distribution networks, to ensure uninterrupted service. These features significantly increase cost.

AI facilities, particularly those focused on model training rather than real-time services, can sometimes accept lower resilience levels, reducing the need for expensive redundancy systems. This flexibility allows for more efficient and potentially lower-cost designs.

A key cost lever is power density. AI data centres often support significantly higher power loads per rack, enabling operators to deploy equivalent or greater IT capacity in a smaller footprint. This can reduce land use and construction time, further improving project economics.

Additionally, purpose-built mega campuses designed for AI model training can realise economies of scale that are harder to achieve in smaller or phased developments. By consolidating infrastructure across multiple interconnected buildings, these campuses offer more cost-effective deployment models.

Turner & Townsend concludes that while investment in data centre infrastructure remains strong, developers and clients must adapt to a bifurcating market. Success will depend on understanding evolving cost structures, mitigating delivery risks, and aligning design strategies with emerging demands from AI workloads.