U.S. Data Center Construction Costs to Rise 13% by 2026, More Than Double the Global Average
nashnova research
US data center construction costs rose 13% year-on-year in 2026, more than double the global average — a land-power-labor squeeze driven by surging AI demand is making every new facility sharply more expensive to build.
How extreme is a 13% jump?
Turner & Townsend's report covers 62 markets and over 500 projects worldwide. Its finding: US data center construction costs rose 13% in 2026 — more than double the global average and roughly three times the pace of the past decade.
This means → The US is not just the world's largest data center market; it is also where costs are inflating fastest. Building the same facility in the US costs far more than the global norm, and the gap is widening.
In plain terms = the rate at which it got more expensive to build a data center a decade ago — multiply that by three for today.
Where is the money going?
The core driver is surging AI compute demand. Power-ready land is in acute shortage, compounded by tight labor markets.
Paul Barry, Turner & Townsend's North American data center lead, notes that compressed schedules force contractors to price in higher risk premiums — rush bonuses, overtime pay, and elevated rates for delivering at abnormal speed.
In plain terms = the faster the client pushes, the more the contractor charges. Materials aren't the main cost driver — the act of rushing is.
How severe is the power bottleneck?
56% of respondents cited power access as the primary delivery barrier. Over half said utility connections typically take one to two years; nearly a third reported lead times exceeding two years.
Regional cost gaps are stark: Silicon Valley runs about $15.01 per watt; Dallas about $10.51 (benchmark: a typical 60 MW design).
This means → Site selection is no longer mainly about cheap land. Whether you can get power — and how fast — is what determines whether a project can proceed at all.
Is there enough labor?
Just over one quarter of respondents said their contractors had sufficient capacity to absorb the backlog of projects under construction.
This means → roughly three in four projects face a workforce shortfall — land, power, and capital may all be in place, but there aren't enough crews to build.
Will costs come down?
Barry expects the pace to slow to 6%–10% over the next year. Many large projects have already broken ground; risk premiums are locked into contracts, and the 2026 surge peak has largely passed.
Over the longer term, PwC projects global data center capex to expand from roughly $800 billion this year to $1.8 trillion per year by 2050. The US is expected to capture nearly half of global AI infrastructure investment.
This reflects a critical distinction: a peaking growth *rate* does not mean falling costs. The demand curve is still climbing, and whether costs can materially decline under sustained high demand is the key variable for AI infrastructure ROI.
市场有风险,内容仅供研究参考,不构成投资建议。
