America's AI data center boom is hitting serious roadblocks. A new analysis from Goldman Sachs shows that roughly 72% of planned data center capacity historically comes online on schedule. But by 2028, only about half of new computing capacity for AI may be ready on time, as shortages of power, chips, materials, land and skilled workers slow expansion.
At the end of last year, the United States had 5,427 operating data centers, according to Stanford University's AI Index Report. Companies announced plans to build another 3,969 facilities. Only 802 are currently under construction.
Goldman Sachs explains that developers often submit applications in multiple regions at once and later choose the site with the best conditions. Columbia Business School real estate professor Stijn Van Nieuwerburgh believes that of the 565 gigawatts of computing capacity now planned by AI companies, only 180 gigawatts may be delivered over the next decade.
The gap between announced and built is wide
JPMorgan estimates 2025 investment in AI infrastructure at $750 billion. But roughly 60% of capacity expected to be ready by the end of 2027 has not yet entered the construction phase. Another 7% of projects where work has already begun are behind schedule.
"Two-thirds of the pipeline is not plausible," Van Nieuwerburgh said.
Even this modest scenario would require roughly $10 trillion in investment. That's 50% more than total spending during the largest previous U.S. investment boom, tied to railroad expansion in the 19th century.
What's blocking construction
Building a data center normally takes 18 to 24 months. Delays are now becoming more common before construction even finishes. Key constraints include:
- Material and chip shortages. Taiwan's TSMC produces nearly all leading AI processors, including Nvidia Blackwell and AMD MI300X chips. Demand has pushed availability and made construction materials harder to source.
- Power grid limits. Data centers already consume about 8% of all U.S. electricity. The American Edge Project forecasts that could rise to 12% by 2028. JPMorgan also found orders for step-up transformers have tripled.
- Energy equipment backlogs. GE Vernova reports its power generator orders have doubled in five years, reaching $36 billion. Since 2020, inflation in transformers and power regulation equipment has been the second-fastest among 47 categories tracked by the Bureau of Labor Statistics.
- Worker shortages. The American Edge Project estimates the U.S. will need an additional 500,000 electricians, 300,000 welders, and 550,000 plumbers to complete the announced projects.
Joe Macejak described the bottleneck: "Some of our customers are developing projects 24/7/365, and contractors move between sites all day, but there is nothing they can do if all the workers are tied up on existing projects."
Permits and public opposition add delays
About ten states have considered a moratorium on data center construction. New York and Texas recently applied temporary restrictions. But Goldman Sachs believes the main obstacle isn't bans - it's the lengthy process of obtaining construction permits.
A Gallup poll found that 71% of Americans do not support development of these facilities. The issue has become part of political campaigns.
Despite the delays, spending on data center construction continues to rise. In June, it increased 7% to $68.3 billion, according to the Census Bureau, which was 46% higher than the same period last year.
Van Nieuwerburgh estimates the cost of building a single modern AI campus at roughly $8 billion. Cleanview reports that 438 unique developers in the U.S. have data center projects.
Minneapolis Fed President Neel Kashkari has noted that such investment levels are increasing inflationary pressure. Data centers cannot be built quickly anywhere in the country: each project requires years of planning, access to electricity, and the involvement of contractors, inspectors, chip manufacturers, and thousands of skilled workers.
Why this matters for real estate and construction professionals
These bottlenecks mean not every announced AI data center project will break ground. For firms that handle site selection, permitting, and infrastructure delivery, the 60% of projects that haven't started construction represent either a massive pipeline - or a major gap between pipe in the sky and actual revenue. Understanding which developers secure power, permits, and workers first will separate viable deals from phantom projects. Those delays also inflate costs across every trade, from electricians to transformer suppliers, which will cascade into broader commercial construction pricing over the next two years.
Your membership also unlocks:
