The AI data center boom is running into a harder constraint than demand: physical supply. New cost, equipment and power data show that building a data center costs far more than it did two years ago, takes far longer to deliver, and increasingly depends on whether a developer can prove it has power before it breaks ground. The data comes from Cushman & Wakefield and industry roundups, and it points to one conclusion: scarcity now sits in the supply chain as much as in the grid.

Construction costs are rising far faster than inflation

Cushman & Wakefield’s 2026 Data Center Development Cost Guide, reported in August by CRE Daily, found that costs rose 21% from the firm’s late-2024 guide. General inflation over the period was roughly 5%. All-in greenfield costs now range from $8.9 million to $23.3 million per megawatt in the United States and Canada, excluding the chips and GPUs needed for AI workloads.

Power infrastructure is the largest single line, at 21% of a greenfield build. Core-and-shell and sitework account for 17%, contingencies for 16%, and site acquisition for 7%. Texas markets are the cheapest, helped by affordable land, labor and contractor networks, while Silicon Valley, Chicago, New York, Northern New Jersey, Toronto and Los Angeles are among the most expensive.

Equipment is the bottleneck

The cost guide tracks equipment prices from December 2021 to early 2026. Switchgear rose 60%, copper 46%, generators and motors 33%, cooling equipment 32% and transformers 29%.

Lead times matter as much as price. The guide puts pad-mounted transformers at 68 to 113 weeks, generators at 60 to 100 weeks and medium-voltage switchgear at 38 to 63 weeks. An industry roundup from Mission Critical Connections last week added that generator lead times of about two years are pushing builders to order equipment before they finalize designs, and that construction materials prices rose 11% in the latest producer price index, the fastest pace since May 2022.

Utilities are screening out speculative projects

Utilities are responding to a flood of power requests by raising the bar. Higher electricity rates and collateral requirements are being used to screen out speculative “phantom” data center projects, according to the Wall Street Journal as summarized in the roundup. Virginia is shifting more risk to developers through utility rate reforms.

Land markets reflect the same shift. U.S. data center land purchases reached roughly $6 billion in the first half of 2026, and access to power now drives land value more than location does, the roundup said.

Demand alone will not carry a project

JLL has warned that strong AI demand cannot protect a project that lacks power certainty, anchor tenants or a credible delivery timeline. That caution matters because the pipeline is enormous. Cushman & Wakefield counts about $2.3 trillion in the North American pipeline, including $492 billion under construction and $1.8 trillion planned or precommitted.

BlackRock CEO Larry Fink added a macro warning last week, telling CNBC that a moratorium on data centers would push the United States into recession. The warning shows how closely financiers now tie the economy to construction continuing.

What the pressure does to competition

Cost and scarcity favor well-capitalized players. Microsoft, Amazon, Google and Meta can pay for equipment early, post collateral and wait out interconnection queues. A smaller developer without a signed anchor tenant faces higher costs, longer waits and tougher utility screening.

That dynamic helps existing capacity. Operators with powered shells already built can lease them at a premium, and equipment makers with order backlogs, such as the generator suppliers whose lead times now stretch past a year, hold pricing power. Developers who locked in turbines, transformers and switchgear early gain an edge over those still negotiating delivery dates.

The same pressure pushes projects toward alternatives. A 1 GW natural gas-powered, behind-the-meter campus won approval from local officials in Taylor, Texas, and Houston firms have formed a partnership to explore powering data centers at sea with stranded natural gas. Both reflect a search for power that does not wait on the grid.

The result is a market that rewards early commitments. A developer that orders equipment before design is final takes on risk, but it also secures a delivery slot that later entrants cannot get. In a market where lead times run past a year, that timing advantage can decide which projects reach completion first, and which tenants get capacity on the schedule they need.

What to watch

Three indicators will show whether the squeeze eases or tightens. First, whether equipment lead times shorten as manufacturers add capacity. Second, whether utility collateral rules thin the pipeline of speculative projects enough to protect the credible ones. Third, whether the cost guide’s next edition shows the 21% rise slowing. Until at least one of those moves, the constraint on AI capacity will be steel, copper and switchgear before it is chips.