The money is not the problem. America’s artificial intelligence infrastructure build-out has capital, land, and political will. What it lacks is electrons, transformers, and approved permits. Six months into 2026, the gap between announced AI data center capacity and what is actually rising out of the dirt has turned into a chasm that threatens to slow the entire sector.
Developers have announced roughly 12 gigawatts of U.S. data center projects for this year, spanning 140 individual sites tracked by industry monitors. Only about five gigawatts are currently under construction. The rest remain stuck in a gauntlet of grid interconnection queues, equipment shortages, and local zoning fights that show no sign of easing. A JPMorgan analysis cited by the Wall Street Journal found that more than 60 percent of capacity planned for 2027 has not even broken ground yet.
The physics of power delivery is the hard ceiling. Goldman Sachs estimates that U.S. data center power demand will climb from 31 GW last year to 41 GW this year, then surge toward 66 GW in 2027. The bank expects only half to 60 percent of near-term scheduled capacity to come online on time. High-voltage transformers, which step down grid power for server racks, now carry lead times of up to five years. Before 2020, those same units shipped in 24 to 30 months. Grid operators like PJM have watched the bottleneck shift from paperwork in interconnection queues to the physical reality of transmission lines and substation buildouts that cannot be expedited by throwing more money at them.
Capital expenditure has never been more abundant, yet cash cannot buy time on a factory floor or override a county board. As one industry voice noted this week:
While consumer tech races ahead with AI features, from smart glasses unveiled at Google I/O 2026 to new smartphone chips, the industrial backbone lags years behind. The power requirements make even the largest mobile batteries look trivial by comparison. A single hyperscale facility can consume as much electricity as a mid-sized city, and utilities simply were not built to absorb dozens of those loads simultaneously.
Local opposition is compounding the technical delays. In Northern Virginia, the nation’s largest data center market, landowners and preservationists have organized to block or stall projects before they reach the construction phase. The American Prospect reported that community pushback has become a structural drag, adding months or years to timelines that were already strained by utility negotiations. Permitting fights once treated as background noise are now primary risk factors, and developers are discovering that a favorable land deal means little if the county refuses to sign off on water rights or substation sight lines.
The result is a widening mismatch between software ambition and hardware reality. Google’s latest Gemini models and competing AI systems grow more demanding with each release, but the buildings that house them face 3-to-5-year waits just to connect to the grid. Unless federal and state regulators streamline permitting and utilities accelerate transmission upgrades, the U.S. risks ceding the pace of AI deployment to regions where infrastructure moves faster. For now, the binding constraint is not intelligence, but electricity.
