SPOTLIGHT

Focus on AI data centers

Data centers specially built to handle the heightened power, cooling and computing demands for AI have become a distinct subsector within the wider market. Our analysis suggests that in the US these ‘AI-ready’ sites carry around a 10% cost premium compared with traditional facilities.

The most advanced AI-specific facilities in the US are seeing cost-per-Watt construction rates of between US$14.00 and US$16.00, significantly above the US$12.08 country average.

The factors below show the main areas that can put upwards and downwards pressure on AI-ready data center construction.

10%

Typical AI construction premium in the US

Key driver	Cost impact	Schedule impact
The location and scale of gigawatt campuses – site selection teams are typically looking for 250MW of IT load+.	Labor shortages and a limited contractor pool capable of delivering at mega scale puts upwards pressure on bid price, particularly in more remote site locations where significant per diem costs need to be absorbed. General ownership costs for the land are also higher.	Contract awards and long-lead procurement have to begin before design is fixed because of the sheer volume of equipment required.
Requirements for higher rack density and liquid cooling – this is the key factor in capital. Some AI-ready facilities operate at 140kW per rack, with the trajectory heading to 1MW, compared to a 2026 US average of 27kW.	Applies a significant premium on capital expenditure, concentrated on the mechanical package.	The commissioning program lengthens due to the requirements for flushing, filtration, and chemical baselining.
Requirement for lower redundancy – as AI training can be paused and restarted so interruption is a less critical issue.	Fewer generators may be needed, and lower expectations of central UPS can reduce the electrical capital expenditure.	Removes some generator lead time and any associated permitting from the critical path.
Rack-level battery storage – installation of energy storage units directly inside or next to server racks to provide immediate backup power directly when needed.	The design change shifts costs around, but ultimately may not impact the total cost directly.	Additional hazard mitigation analysis is required because the batteries are now spread throughout the main server room rather than isolated in a dedicated, fire-rated UPS room.
On-site power generation – to reduce reliance on gaining grid connections, a critical part of planning and design.	Capital expenditure on power generation infrastructure such as turbines and engines could be significant. Already demand is pushing up prices, with some suppliers reporting turbine orders in 2026 are priced 10–20% higher per kW than in late 2025.	Removes the risk of a long grid queue, but replaces this with risks from planning and permitting, as well as eventual interconnection to the grid.

Key factors shaping US AI facility costs and schedules

Department of Energy regulations. New rules could restrict foreign energy equipment, delaying large data center projects and energization.
Electric Reliability Council of Texas Batch Zero. Queue review results will indicate how much new data center load the Texas grid can support.
Mechanical, electrical and plumbing labor inflation. It will be important to watch whether mechanical, electrical, and plumbing wage escalation remains around 7%, continuing to outpace broader construction inflation.

Key takeaways for markets beyond the US

While the cost premium shown in our data is built on some elements that are unique to the US, there are some broader points of relevance to other regions:

Contractor capacity The extent of the premium for AI-ready facilities will depend in part on the number of large contractors available locally capable of delivering these projects at scale.

Redundancy requirements Redundancy requirements, and therefore how much they impact the premium, will depend in some cases on local regulations and customer requirements.

Power availability The costs of on-site generation can be mitigated if a project is built in an area with more abundant or flexible grid connections.

Lessons from commissioning Lessons learned from testing and commissioning US AI facilities through 2027 could help improve the design and construction of similar facilities in other regions.

Workforce welfare US data center projects are placing greater emphasis on transport, accommodation and workforce welfare when delivering large-scale facilities in remote locations.