The electric grid in the United States is currently facing an unprecedented surge in data center proposals, with experts indicating that only a small portion will actually materialize. Industry insiders suggest that interconnection requests exceed actual construction by a factor of five to ten, creating significant challenges for utilities and grid operators in their planning efforts.
The disparity in AI power consumption forecasts illustrates this challenge. While RAND Corporation projected 347 GW of AI-sector power usage by 2030, Schneider Electric’s analysis suggests much lower figures, ranging from 16.5 GW to 65.3 GW. This wide variation highlights the difficulty in planning for future power needs.
These speculative requests, dubbed “phantom data centers” by industry analysts, are causing substantial complications. Even major tech companies like Microsoft, Meta, Amazon, and Google routinely propose multiple projects knowing many won’t proceed. Microsoft recently withdrew up to 2 GW of capacity reservations, while Tract canceled a major Phoenix development amid community opposition.
The situation is further complicated by developers increasingly turning to behind-the-meter power generation. Notable examples include Elon Musk’s xAI facility in Memphis, which operates 35 gas turbines independently, and Liberty Energy’s planned 1 GW off-grid generation near Pittsburgh.
Utilities are grappling with these challenges in various ways. Great River Energy, a Minnesota-based cooperative, has implemented fees for evaluating large-load requests to protect existing customers. The utility has received numerous interconnection requests but remains cautious about committing resources to potentially speculative projects.
In Minnesota alone, utilities have received proposals for at least 11 data center campuses since 2020, including projects from tech giants Amazon, Microsoft, and Meta. However, only Meta’s project had begun construction as of early 2025, highlighting the gap between proposals and actual development.
Recent research by the Electric Power Research Institute reveals utilities’ skepticism about proposed data center loads. In a September 2024 survey of 25 major utilities, while 48% expected data centers to represent at least 10% of peak load by 2030, none anticipated actual five-year usage exceeding 35% of peak load, even in areas with the highest concentration of requests.
Solutions are emerging across the industry. Some utilities are implementing standardized processes for large-load interconnections, while others require larger upfront financial commitments. In Virginia, the nation’s largest data center market, three utilities have proposed new rate classes specifically for data centers. Dominion Energy and Appalachian Power would require minimum load commitments of 60% and 80% respectively, while Rappahannock Electric Cooperative’s proposal includes collateral requirements and infrastructure upgrade responsibilities.
Industry experts suggest more comprehensive solutions may be needed. Former Federal Energy Regulatory Commissioner Allison Clements and former Meta executive Peter Freed advocate for a nationwide
standardized process to reduce speculative requests and accelerate interconnection timelines. Their proposed framework includes standardized queues, transparency in project status, and progressive fee structures.
However, some experts, like former Texas Public Utility Commissioner Karl Rábago, argue that current solutions may be insufficient. Rábago suggests that developers will continue to secure queue positions as long as the associated costs remain lower than the potential benefits, proposing instead a “reverse auction” system to determine which data centers require minimal incentives for grid connection.