The Stargate Project: Can the U.S. Grid Support $500 Billion in AI Infrastructure?
Virginia's electricity generation is up 16.4% year-over-year. Texas wind drops to 14% of output during peak summer heat. Industrial power costs 40% more in Virginia than in Texas. The data tells a specific story about where Stargate's buildout is already straining the grid, and where it will next.
In January 2025, OpenAI, SoftBank, Oracle, and MGX announced a joint venture to invest $500 billion in AI computing infrastructure across the United States by 2029. The flagship campus in Abilene, Texas had two buildings operational by September 2025 with NVIDIA GB200 racks already running early training and inference workloads. Five additional U.S. sites were announced simultaneously, bringing planned capacity to nearly 7 gigawatts and over $400 billion in committed investment.
At 1.59% of current U.S. GDP, this is not a technology announcement. It is a macroeconomic event with direct, measurable consequences for the electricity grid. This report quantifies those consequences using exclusively federal agency data.
The U.S. grid can physically accommodate Stargate-scale load nationally, but not instantaneously, not uniformly, or cheaply. The Abilene flagship site is already operational and drawing power from ERCOT, the only major U.S. grid with no external interconnection. Virginia, the most data center-dense state in the country, is up 16.4% year-over-year on a grid that is 63.6% natural gas with essentially no wind capacity. Texas wind (the grid's largest renewable source) drops to 14% of output during peak summer heat, forcing natural gas to 56% of generation exactly when demand is highest.
The 40% industrial electricity price gap between Texas and Virginia is the economic logic behind the Abilene siting decision. For a 1 GW data center campus, that differential is $199 million per year in operating cost savings at prevailing industrial rates. At 5 GW portfolio scale, it approaches $1 billion annually. But it places the facility on the most grid-constrained system in the dataset: a system that has now recorded three consecutive summers approaching its generation ceiling, before a single Stargate campus reaches full operational scale.
BEA national accounts data shows private fixed investment in IT equipment and software rising from $1.104 trillion annualized in Q1 2023 to $1.466 trillion in Q3 2025, a 32.8% increase that is the fastest rate of growth in the dataset and predates the Stargate announcement. The grid demand impact from this investment will compound over the next 24 to 36 months as facilities across the sector come online.
Henry Hub natural gas spiked to $7.72/MMBtu in January 2026: 157% above the March 2024 low of $1.49 and nearly double December 2025's $4.26. The dataset shows a moderate positive correlation (r = 0.31) between gas prices and industrial electricity prices with a one-month lag. Virginia and Texas grids meet every increment of new AI demand with natural gas. Facilities with long-term power purchase agreements are insulated from this volatility. Facilities exposed to market rates are not.
What This Report Contains: 18 Pages, 25 Federal Data Series
- 01 Executive Summary: Key findings and bottom line FREE
- 02 National Baseline: Monthly U.S. generation, full 2025 bar chart 🔒
- 03 Virginia: 12-month generation growth chart, commercial sales data 🔒
- 04 Virginia Fuel Mix: Full breakdown by source with grid dependency analysis 🔒
- 05 Texas / ERCOT: Three-year generation ceiling analysis 🔒
- 06 Texas Fuel Mix: Monthly wind vs. gas seasonality table, all 12 months 🔒
- 07 Cost Equation: Industrial price comparison, state vs. national, 36-month trend 🔒
- 08 Household Impact: Residential and commercial prices across all three geographies 🔒
- 09 Investment Wave: Full 11-quarter BEA IT investment table, GDP context 🔒
- 10 Volatility Risk: Full 36-month Henry Hub price dataset, correlation analysis 🔒
- 11 The Verdict: Three questions answered, bottom line 🔒
- A1 Data Appendix: All 25 series documented with source, frequency, coverage 🔒
- A2 Endnotes: Full citation list with URLs for all external sources 🔒
Access the Full Report
Subscribe for full access to this report and every Silicon Curve publication, or purchase this report once at no ongoing cost.
Already a subscriber? Sign in