The AI data center boom is not materializing as fast as the headlines suggest. Despite $750 billion in AI infrastructure investment this year, only about half of the AI computing capacity scheduled to come online by 2028 will be operational on time, according to Goldman Sachs. Data centers typically take 18 to 24 months to build, but timelines are stretching as delays compound.
The gap between plans and reality
The planned expansion is enormous. The US had 5,427 data centers at the end of last year, according to Stanford's AI Index. Aterio counts 3,969 announced plans for new US data centers, but only 802 are under construction.
Developers often file applications for multiple sites and then pick the most viable one, Goldman Sachs said. About 60% of the capacity scheduled for completion in 2027 has not yet begun construction, and roughly 7% of projects that did start have been delayed. Columbia Business School real estate professor Stijn Van Nieuwerburgh expects only 180 gigawatts of the 565 gigawatts of planned computing capacity to actually be built over the next decade. Two-thirds of the development pipeline is unrealistic, he said.
The numbers add up to roughly $10 trillion in potential investment, 50% more than the 19th-century infrastructure expansion. Demand has outrun the industry's ability to deliver. Yet spending keeps climbing: data center construction rose 7% in June to $68.3 billion, up 46% from a year earlier, according to the Census Bureau. A single modern AI campus costs about $8 billion, Van Nieuwerburgh said.
Materials, chips, and electricity
Construction materials are hard to source as demand surges. Even when materials arrive, the chips that fill these buildings are in short supply. Taiwan's TSMC manufactures virtually all leading AI chips, including Nvidia's Blackwell and AMD's MI300X, making it a single point of dependency in the global supply chain, according to Stanford's AI Index.
Electricity is another bottleneck. Data centers already consume about 8% of US electricity, and that could rise to 12% by 2028, according to the American Edge Project. Many AI companies are building on-site electricity generation, but waiting times for step-up transformers have tripled, according to JPMorgan. GE Vernova, the largest gas turbine maker, has seen generator orders double to $200 billion over five years. Transformer lead times and labor availability are the supply chain signals that AI for Operations teams track to separate viable projects from announcements.
Labor shortages and public opposition
Building the proposed data centers would require 500,000 electricians, 300,000 welders, and 550,000 plumbers, according to the American Edge Project. The workforce is not there.
"Some of our clients are developing around the clock, seven days a week, 365 days a year, and contractors are moving throughout the day, but there is nothing they can do if the entire workforce is occupied with existing projects," said Joe Macejak.
Public opposition is growing alongside the labor crunch. Gallup found 71% of Americans oppose AI data centers. About a dozen states have proposed moratoriums, and New York and Texas have imposed temporary bans. Goldman Sachs said obtaining construction permits remains a bigger hurdle than outright bans.
Why this matters for real estate and construction
For developers and contractors, the gap between announced projects and shovels in the ground is a market signal. Professionals working in AI for Real Estate & Construction should treat announced data center capacity as speculative until financing, permits, and electricity agreements are locked in. The more reliable opportunities may be in retrofitting existing industrial sites with available electricity and permitting, rather than chasing greenfield projects that may never break ground.
The cost of delay is real. "It is very difficult to accurately estimate the timelines for such large construction projects, and what often happens is that we become heavily burdened with debt, causing many of these investments to fall apart," Van Nieuwerburgh said. Minneapolis Federal Reserve President Neel Kashkari said last week that data centers are now contributing to inflation.
Your membership also unlocks:
