> ## Content Index
> Fetch the complete content index at: https://millionaire-insiders.ghost.io/llms.txt
> Use this file to discover other available public pages before exploring further.

# The AI Buildout Became an Electricity Story, and the Bill Is Being Split in Public
- URL: https://millionaire-insiders.ghost.io/the-ai-buildout-became-an-electricity-story-and-the-bill-is-being-split-in-public/
- Published: 2026-09-07T11:30:34.000Z
- Updated: 2026-09-07T11:30:33.000Z
- Author: Daniel Whitmore

Washington State's utility regulator held a workshop last month on what to do about large power users, and it is drafting rules that would let utilities demand mandatory load reductions during shortages, longer contracts and surcharges from data centers. A draft policy statement is due this fall. Several towns in the state have already passed temporary moratoriums on new large data centers.  
  
That is a long way from an earnings call, and it is exactly where the AI trade is being decided now. The constraint on the buildout is no longer chips or capital. It is electricity, and the argument over who pays for it has moved into rate cases and state legislatures.

### The demand numbers are large enough to move a national average

US data centers used about 176 terawatt hours of electricity in 2023, roughly 4.4% of the country's total consumption. Lawrence Berkeley National Laboratory's projections put that between 325 and 580 terawatt hours by 2028, or somewhere from 6.7% to 12% of national use. The width of that range is itself the story, because grids have to be built for the top of it.  
  
Texas offers the clearest picture. ERCOT expects peak summer demand approaching 145 gigawatts by 2031, against 85 gigawatts in 2024, with data centers accounting for roughly 32 gigawatts of the increase. Utilities cannot serve that with existing capacity, which means new generation, new transmission and a decade of capital spending recovered through rates.  
  
The cost is already visible in wholesale markets. In the PJM region, which covers Northern Virginia, Monitoring Analytics attributed about $9.3 billion of capacity costs in the 2025-26 auction to data center demand, putting those costs far above where they would otherwise have been. Dominion Energy in Virginia proposed its first base-rate increase since 1992, adding several dollars a month to the typical household bill.

### Why this is a macro variable and not just a utility story

**Whitmore's Watchlist:** 
**XLU** (Utilities Select Sector SPDR Fund): Broad exposure to the regulated utilities that recover this capital spending through approved rates.  
**CEG** (Constellation Energy Corporation): The largest US operator of nuclear generation, selling into exactly the round-the-clock demand data centers want.  
**GEV** (GE Vernova Inc.): Turbines, grid equipment and electrification hardware, the supply chain behind new generation and transmission.

The July FOMC minutes listed an AI-related demand surge alongside tariffs and supply disruptions as a reason inflation has stayed elevated, with total PCE at 4.1% in May. Electricity is a component of the consumer price basket and an input cost for nearly everything else, so a structural bid for power is not a sector curiosity. It is a small, persistent upward force on the same inflation number the Federal Reserve is deciding about on September 15 and 16.  
  
Regulated utilities occupy an unusual position in this. They are traditionally the defensive, rate-sensitive corner of the market, which means higher long yields normally hurt them. Here the same environment is handing them the largest load growth in decades, which supports the rate base their earnings are built on. Those two forces are pulling in opposite directions, and which one dominates depends heavily on what regulators approve.

### The regulatory question is the investable one

The rules being written now decide whether data centers pay the full incremental cost of the capacity they require, or whether some of it lands in general rates. Special large-load tariffs, minimum take-or-pay commitments and interruptibility clauses are the specific mechanisms, and they vary state by state.  
  
There is a reliability dimension too, and it is not theoretical. In one 2024 incident in Northern Virginia, a voltage fluctuation caused roughly sixty data centers to disconnect from the grid simultaneously, leaving the system with a sudden power surplus of about 1,500 megawatts. Loads that large behaving in unison is a new engineering problem, and solving it costs money that someone will be billed for.  
  
For a utility, a favorable large-load framework converts an operational headache into decades of recoverable investment. An unfavorable one leaves it carrying the capital cost of capacity it may not be able to bill for at the same time as residential customers and their elected representatives get louder about their bills.  
  
**Whitmore's Take**: The AI capital cycle has quietly become a regulated-utility story, and regulated outcomes move slowly and in public. Reading the rate cases is a slower way to follow this theme than watching chip earnings, and probably a more informative one.

![](https://storage.ghost.io/c/e9/10/e9109ad9-55f9-4e96-a078-46af25115156/content/images/2026/09/ai-power-demand-2.jpg)

*Written by Daniel Whitmore* 
*Millionaire Insiders*