Report: Over 70% of U.S. Data Center Power Applications May Never Materialize
Nashnova编辑部
US data centers have filed 1,066 GW of power applications, but grid operators expect only about 28% to materialize — a flood of 'phantom applications' is choking approval pipelines, threatening America's AI buildout pace, and potentially shifting the cost of misplanned capacity onto ordinary ratepayers.
1,066 GW of applications — why will less than a third become real?
Wood Mackenzie's latest forecast: of 1,066 GW in power applications filed by US data centers, grid operators expect to fulfill only about 28%. More than two-thirds will not convert into actual load.
This means → the total application volume equals 83% of all US utility-scale generation capacity at the end of last year — yet most of it is paper demand, not real load.
The core driver is "phantom applications" — developers file the same project with multiple utilities, pick the best approval, and abandon the rest. Rapidan Energy Group estimates only 20–30% of the power developers seek will serve completed projects.
Which regions have the most inflated queues?
Rystad Energy assessed PJM — the largest US grid, covering roughly 67 million customers from Illinois to Virginia — and found about half the applications credible.
In Texas, where data-center growth is fastest, the credible share on the main grid drops to roughly 14%.
In plain terms = Texas has the biggest queue and the biggest bubble — roughly one in seven applications is serious.
What happens when approval pipelines jam?
The sheer volume has overloaded grid review processes and stretched timelines. Halcyon co-founder Alex Klaessig warned: "If we can't bring these data centers online quickly, we risk missing out on the gains AI will deliver."
To stem the flood, several utilities now demand large deposits and near-perfect credit ratings. This reflects a shift from open-door intake to active screening.
This means → smaller developers are hit hardest. They typically front construction costs, then sell finished facilities to well-funded AI firms — and now some cannot even qualify to apply. Hut 8 SVP Brad Richter said most utilities have "stopped taking on new business."
Who pays the bill when plans go wrong?
Grid upgrades are typically recouped through higher electricity rates. This means → if large numbers of projects fall through, the cost of capacity already built may land on ordinary ratepayers.
In plain terms = capacity reserved for phantom projects doesn't disappear for free — the upgrade bill gets spread across everyone's power costs.
Texas halts approvals — how big is the impact?
ERCOT — Texas's grid operator — tracks roughly 474 GW of interconnection applications, about 90% from data centers. That total exceeds the system's historical peak demand by more than five times.
On August 3, Governor Greg Abbott ordered an audit of every data-center project seeking access to the state's main grid; approvals were immediately paused. BloombergNEF estimates the move puts $13 billion in industry revenue at risk.
This means → roughly one-fifth of the entire US data-center construction pipeline could be delayed in the medium term — one state's regulatory action is enough to reshape the national AI-infrastructure timeline.
Whose projects will actually get built — and who is most exposed?
Wood Mackenzie analyst Caitlin Connelly noted that many applications come from companies that have never built a large-scale data center before. These first-time entrants tend to file oversized projects with far lower completion odds than established players like Google and Microsoft.
This reflects a reality now becoming central to the US AI buildout: queue credibility screening is the key variable determining whether infrastructure can accelerate.
In plain terms = the grid is moving from a first-come-first-served line to a qualification-based gate — not "who queued first" but "who can actually build."
Content is for reference only, not financial advice.