US AI data center construction faces significant delays despite strong investment plans
Although billions are slated for AI data center projects, only a fraction are under construction, with material, power, labor shortages and permitting hurdles causing widespread delays.
Construction of artificial-intelligence data centers in the United States is lagging behind ambitious rollout plans. A Gallup poll found that 71% of Americans oppose new AI data centers, and several states have introduced or enacted bans, but the primary obstacle appears to be the difficulty of building the facilities.
Goldman Sachs estimates that only about half of the AI computing capacity scheduled for activation by 2028 will be operational on time, and JPMorgan reports that roughly 60% of capacity slated for completion in 2027 has not yet begun construction. Of the 3,969 new AI data centers announced by companies, just 802 are currently being built.
Industry analysts note that developers often file multiple site applications and later select the most viable, leading to an inflated pipeline. Columbia Business School professor Stijn Van Nieuwerburgh expects only 180 gigawatts of the planned 565 gigawatts of AI computing power to be realized over the next decade, describing two-thirds of the pipeline as “implausible.”
Key factors contributing to the slowdown include shortages of building materials, limited availability of advanced AI chips—particularly from Taiwan’s TSMC—and growing strain on the electrical grid. The American Edge Project projects that data centers could account for up to 12% of U.S. electricity use by 2028. Delays in obtaining step-up transformers and a surge in prices for power-regulating equipment have further hampered progress.
Labor constraints also pose a challenge; the sector would need hundreds of thousands of additional electricians, welders and plumbers, a need not fully met by recent immigration policy changes. “Some of our clients are developing 24/7/365, and contractors are moving around all day, but there’s nothing they can do if all the labor is tied up in existing projects,” said Joe Macejak, head of Marsh Risk’s U.S. property digital infrastructure business.
Despite these setbacks, spending on data-center construction rose 7% in June to $68.3 billion, a 46% increase from the previous year, according to a Census Bureau report. A single state-of-the-art AI campus can cost around $8 billion, and 438 developers currently have projects in the United States, per Cleanview.
“The timing is very hard to get right with these big buildouts, and often what ends up happening is we get over-excited and accrue too much debt and then a bunch of these investments go bust,” Van Nieuwerburgh added.