AI’s Power Demands Reshape Utility Gatekeeping

AI operations require staggering amounts of electricity, creating a bottleneck as demand outpaces grid capacity in the United States. This shortage is transferring leverage from semiconductor companies like NVIDIA to utilities and grid operators who control power distribution.

In June, the Electric Reliability Council of Texas (ERCOT) revamped its process for admitting large power users, addressing a backlog of data centers, cryptocurrency mines, and industrial facilities competing for megawatts. Simultaneously, Albany, New York lawmakers considered a one-year moratorium on new large-scale data centers, potentially making it the first state to halt construction outright.

Companies building frontier AI models are encountering obstacles related to land availability, generation capacity, water resources, high-voltage transformers, and local regulatory approval. Utilities benefit from this demand surge, collecting payments regardless of which company wins the AI competition while earning regulated returns on grid upgrades.

The conversation has fundamentally shifted from software and GPU supply to industrial infrastructure. Goldman Sachs forecasts U.S. data center power demand rising from 31 gigawatts in 2025 to 66 gigawatts by 2027, increasing data centers’ share of peak summer demand from 4.1% to 8.5%. However, only 50-60% of planned capacity will likely materialize on schedule due to delays and cancellations.

The International Energy Agency expects data center electricity use to double by 2030, with AI-focused facilities tripling their consumption. This growth faces bottlenecks in supply chains for critical equipment and lengthy grid connection processes.

Texas exemplifies this gatekeeping role. Senate Bill 6 requires large customers to pay their own interconnection costs and maintain reserve capacity for emergencies, imposing a $50,000-per-megawatt fee to prevent speculative applications. Nearly 200 large users sought 438 gigawatts in the first quarter of 2026—over five times the state’s current draw.

New York’s proposed pause weighs AI data center expansion against household electricity costs, water usage, and grid reliability. Power has become a rationed resource, with utilities and grid operators holding the strongest negotiating position.

Bitcoin mining pioneered this type of infrastructure competition by using flexible, interruptible power during surplus periods. However, AI requires steady, always-on power backed by long-term commitments and national competitiveness arguments. When BlackRock predicted in January that AI data centers could consume 24% of U.S. electricity by 2030, it signaled the end of cheap-power negotiations with miners.

Utilities now arbitrate between competing demands, with regulated operators earning returns on approved projects while independent power producers sell into tighter, higher-priced markets. Ratepayers may absorb infrastructure costs unless regulators specifically allocate them to large commercial users.

The EIA expects U.S. power consumption to set records in 2026 and 2027, with residential prices already rising 5% in 2026, particularly along the East Coast. Despite AI’s promise of software abstraction, electricity has become the critical constraint determining which companies can scale, who gets priced out, and who profits from the ongoing expansion.