A massive infrastructure project in West Texas will generate more electricity than all of Chicago uses - and none of it will flow through the public grid. Welcome to the shadow power grid, where AI companies are becoming their own utilities.
The GW Ranch Project
Pacifico Energy’s GW Ranch in Pecos County, Texas, represents the most ambitious private power project in American history. The Texas Commission on Environmental Quality approved a 7.65-gigawatt gas-fired power campus, combined with 1.8 GW of battery storage and 750 MW of solar capacity.
For context: that’s enough to power roughly 5 million homes. The entire state of West Virginia uses less electricity than what this single facility will generate.
Construction begins in Q1 2026, with first power delivery scheduled for Q1 2027. The project can scale to over 5 GW by 2031 without requiring additional regulatory approvals.
GW Ranch won’t connect to the public power grid at all. Dozens of airplane-hangar-sized warehouses packed with AI computing hardware will operate entirely on self-generated electricity.
Why Companies Are Going Off-Grid
The American power grid wasn’t built for AI data centers. PJM Interconnection, the largest U.S. grid operator serving 65 million people across 13 states, projects it will be six gigawatts short of its reliability requirements by 2027.
Connecting new data centers to the grid means waiting years for interconnection approval. Building your own power plant is faster.
The Trump administration has encouraged this approach. The vision: “AI companies become power companies” and build their own generation alongside new data centers.
The result is a parallel energy infrastructure emerging across the country. Projects already approved by state regulators could power all of New York City several times over.
The Bipartisan Response
Senators Josh Hawley (R-Mo.) and Richard Blumenthal (D-Conn.) introduced the GRID Act (Guaranteeing Rate Insulation from Data Centers) in February 2026 - the first bipartisan federal legislation targeting data center power consumption.
The bill’s core requirements:
Off-grid mandate: Data centers with power demand of 20 megawatts or more cannot draw electricity from the public grid unless they operate via off-grid sources or prove they don’t increase rates for other customers.
Certification requirement: Existing data centers get 10 years to either transition off-grid or obtain a federal “Zero Rate Effect Certificate” proving they don’t raise consumer electricity prices.
Transparency rules: Utilities and operators must publicly disclose any agreements for data center service, including special tariffs, discounts, tax incentives, cost-sharing deals, or subsidies. Companies must estimate total savings and disclose any financial ties to utilities.
Penalties: Violations carry civil fines of at least $1 million per day.
Hawley framed it directly: “American families should not have to shoulder the burden of rising electricity costs” from data centers. Blumenthal called it stopping tech companies’ “AI-driven drain on family’s pocketbooks.”
The Critics’ Case
The Center for Data Innovation argues the legislation misdiagnoses the problem. Their main objections:
Flexible demand, not inelastic: Large portions of AI training workloads can pause, slow down, or shift to off-peak hours without significant losses. The bill treats data centers as purely extractive consumers rather than potential grid partners.
Market design, not corporate behavior: The real issue is outdated grid management, not data center malice. Treating presence in the electricity market as inherently extractive “confuses poor incentives with malintent.”
Texas as alternative model: Texas Senate Bill 6 conditions grid access on automated curtailment technology, allowing data centers to “surgically ramp down” consumption during emergencies while maintaining operations through on-site microgrids. This approach enables faster interconnection while creating financial incentives for grid cooperation.
Community Backlash
The projects are sparking resistance across the country.
Palm Beach County, Florida: A proposed 200-acre data center complex was postponed after residents cited noise from cooling towers, heavy water use, and pollution concerns.
Austin, Texas: Multiple proposed data centers seek more than five gigawatts of power - exceeding the city’s entire peak load capacity.
Davis, West Virginia: The Monarch Compute Campus will initially generate enough electricity to power 1.5 million homes, with plans to later quadruple output - several times the total electricity consumption of West Virginia residents.
The political opposition spans the spectrum. Both Bernie Sanders and Ron DeSantis have spoken against data center expansion, indicating bipartisan concern at the state level as well.
The Environmental Cost
These off-grid projects run primarily on natural gas. The carbon footprint is substantial.
Approved off-grid data centers will increase U.S. emissions of carbon dioxide and other air pollutants. The irony: companies building AI systems to solve climate problems are simultaneously building fossil-fuel power plants to run those systems.
The alternative - connecting to increasingly renewable-powered grids - would require waiting for interconnection that may take years. Speed is winning over sustainability.
Who Pays
The fundamental question behind the GRID Act: should regular customers bear any costs when data centers outbid utilities for equipment, leaving other ratepayers to cover maintenance for older infrastructure?
Critics of the legislation argue that forcing data centers off-grid entirely eliminates any chance of them becoming useful grid partners who can reduce consumption during peak demand.
Supporters argue that voluntary cooperation hasn’t materialized and legislative pressure is necessary.
More than 300 state data center bills have been filed across 30+ states in the first six weeks of 2026 alone. Virginia, Georgia, and Oklahoma are reconsidering the tax incentives that previously attracted these facilities.
What This Means
The AI industry’s energy demands have outgrown the infrastructure designed to deliver electricity. The response is splitting into two tracks:
Track one: Build private power and operate entirely outside the grid system. This is faster but raises emissions and sidesteps public utility oversight.
Track two: Force data centers to prove they’re not hurting other ratepayers, potentially slowing AI infrastructure development but protecting consumers.
The GW Ranch project and the GRID Act represent these competing visions. One says AI companies should become their own utilities. The other says that if they want grid access, they need to prove they’re not making everyone else’s bills more expensive.
Both agree on one thing: the current approach isn’t working.