Data center project count nearly triples; power is the bottleneck

Construction Dive's analysis of Cleanview data shows the number of planned and operating US data center projects nearly tripled between March and August 2026, even as state moratoriums and public pushback made data centers a political target. The surge is best read as a queue of developers betting they can secure reliable power before a regulator or a neighborhood coalition stops them.

Money is not the constraint. J.P. Morgan estimates the largest hyperscalers will spend a combined $697 billion on data center capital expenditures in 2026, and Meta, Amazon, Google and Microsoft opened the AI buildout, with colocation developers following. ABC chief economist Anirban Basu says the sector has been propping up nonresidential construction for most of the year, which pushes the pressure point down the timeline from raising capital to actual delivery.

Construction Dive's review puts Microsoft and Amazon at the top of the project list, with Google and Stack Infrastructure planning scores of their own. The geography is familiar—Virginia remains the data center capital, Texas and Georgia lead the South, and Ohio and Indiana anchor the Midwest—but the less familiar shift is inside the size of what those states are being asked to host.

Just a couple of years ago, a 100-megawatt data center would have ranked among the largest projects on the market, David Guarino, head of global data center and tower research at Green Street, told Construction Dive in April. Today, projects above 1,000 megawatts set the benchmark, and the jump marks the difference between a building permit and a grid-level event: a developer no longer picks a site and then asks for power, because the power assessment comes first and the site follows.

Michael Beadle, senior vice president of operations at Minneapolis-based Ryan Cos., put the risk in contractor terms: "The success of a project is tied to securing a site with access to a significant amount of reliable power, which requires strategic and lengthy relationships with utility partners." State restrictions have since narrowed the search; Trent Cotney, a partner at Adams & Reese, told Construction Dive that newly imposed regulations have made site selection a top priority when pursuing these awards.

This publication has argued that grid access is becoming the underwriting variable, set by customer class and budget line rather than queue position. The Cleanview numbers are a direct national stress test of that argument: a pipeline that triples in five months functions as a queue of land positions and utility applications, each with a different chance of surviving contact with a state regulator or a local opposition group.

Some developers are already bypassing the ordinary market for power: the Department of Energy's $1.9 billion loan to restart the Duane Arnold nuclear plant turns Google's PPA into infrastructure, as this publication has reported. That transaction is a separate asset class from the construction pipeline, but it is the logical end point of the same constraint—when the queue is too long, a hyperscaler's balance sheet gets converted into federal financing and the power project gets built first.

The near-tripling in project count will not translate one-for-one into completed data centers, because power constraints, state restrictions, and permitting delays all subtract from the total. What remains is the portion of the pipeline that controls something more durable than an announcement: a utility relationship, a signed interconnection position, a site cleared for construction. Those attributes—not the project count—are what separate digital infrastructure from a land option.

The next question for anyone underwriting this cycle is which projects already have power committed in front of them. Construction Dive's data counts projects; the figure that now matters is megawatts with a committed source behind them.