The AI Data Center Boom Faces a New Reality Check

Since a May 2025 Utility Dive headline warned that only a fraction of proposed data centers would be built as utilities grew wary, industry friction has intensified. Community opposition has emerged as a major drag on development, and by June the energy advisory firm Relae reported more than $170 billion of proposed AI data center capacity had been blocked, withdrawn or stalled since January 2024. At the same time, construction labor shortfalls, long lead times for critical electrical equipment, constrained power in some markets and uncertain end-user demand for AI tools are adding to delays.

Major market analysts nevertheless still project substantial load growth from hyperscalers. Goldman Sachs estimated roughly $581 billion in U.S. hyperscaler spending on AI infrastructure this year and expects U.S. data center power demand to more than double from 2025 levels to about 66 GW by 2027, though it also said only about half of near-term capacity is likely to come online on schedule amid cancellations and delays. The Electric Power Research Institute has forecasted that data centers could account for 9% to 17% of U.S. electricity demand in 2030 and as much as 20% by 2035.

Utilities and regulators are responding by tightening terms for large new loads. Many have adopted or proposed large-load tariffs and contractual requirements — including minimum contract lengths, collateral, upfront study fees, exit charges and ramp schedules — to deter speculative interconnection requests and ensure that grid upgrades are paid for by those who need them. Some states are also offering incentives for developers that bring their own generation, accept flexibility, or align with state clean-energy and economic-development objectives.

The dynamics are especially acute in Texas. The state paused new data center interconnections while auditing a queue totaling 474 GW, more than five times the grid’s recent summer peak of about 90 GW. ERCOT and the Public Utility Commission of Texas have said the forecasted load is likely higher than what will materialize and have moved to refine forecasting and interconnection protocols. In June the PUCT adopted a cluster study framework for prospective loads above 75 MW to streamline sequential studies; ERCOT expects to publish a transmission plan for the initial "Batch Zero" cluster in late 2027. Texas has also enacted gating conditions and requirements under Senate Bill 6 to discourage speculative requests and clarify who bears upgrade costs.

Industry participants and officials say clearer answers on which projects are real, and when they will energize, are needed to avoid either underbuilding or overbuilding the grid — outcomes that could raise costs or risk reliability for other customers. The combination of community pushback, tighter regulatory gates and persistent logistical bottlenecks is increasing uncertainty about the scale, timing and location of the next wave of AI data center load.