A wave of speculative AI data center projects is muddying the outlook for US electricity expansion, raising the risk that utilities will pour money into new capacity that never gets used while genuinely needed projects wait for grid connections.
What are phantom projects?
The problem, described as a 'phantom project' issue, is that many announced AI facilities are not backed by firm commitments. They may be land options, early-stage feasibility studies, or proposals that never secure financing. But they still show up in utility demand forecasts and grid interconnection queues.
For grid planners, the challenge is separating serious proposals from those that are little more than placeholders. A project that looks real on paper can tie up resources for years before it quietly disappears.
Why the grid is vulnerable
AI data centers are power-hungry, and the expected surge in demand has utilities rushing to build new transmission lines and power plants. But the uncertainty around which projects will actually be built makes planning difficult. If utilities base their decisions on phantom demand, they risk building infrastructure that sits idle.
The interconnection queue, where projects line up for grid studies, is especially sensitive. Each request, real or not, consumes time and engineering resources. When speculative projects clog the queue, they push back the timeline for projects that are ready to move forward.
The cost of overbuilding
Overbuilding means higher costs for ratepayers. New power plants and transmission lines are expensive, and if they're built for demand that never arrives, those costs get spread across customers. It also diverts resources—steel, concrete, labor, and capital—away from projects that could actually meet real needs.
The risk is not just wasted money. It's that the grid becomes less reliable because investments are misallocated. A system built for the wrong future is one that may struggle to handle the one that actually arrives.
Delays for real projects
Real projects suffer in two ways. First, they wait longer for grid connection studies because the queue is backed up with speculative requests. Second, they may face higher costs if utilities have already committed to building capacity for phantom demand, leaving less room in the budget for what's actually needed.
The problem is not new, but the scale of AI-related announcements has made it harder to ignore. Utilities and grid operators are left to guess which projects are real, and the cost of guessing wrong is growing.
The next few quarters will show whether grid operators can tighten their forecasting methods and separate real demand from phantom projects before committing billions to new infrastructure.
