THE GIGAWATT CONSTRAINT
The constraint on artificial intelligence in 2025 is no longer silicon. It is the grid.
While investors focus on GPU shipments, a more critical bottleneck has emerged in the physical infrastructure of the United States. Data center power consumption is projected to grow 160% by 2030, according to Goldman Sachs.
The major tech capitals—from Northern Virginia to Silicon Valley—are hitting a physical wall. The existing utility grid cannot support the gigawatt-scale demands of the next generation of AI model training without risking widespread blackouts.
This structural deficit is forcing a geographic realignment of capital. The "Smart Money" is leaving the coastal grids and moving to the American Heartland.
The "Base-Load" Crisis
The issue is physics. AI clusters require "base-load" power—energy that runs 24/7, regardless of weather. Wind and solar, while popular, are intermittent. They cannot guarantee the uptime required for Nvidia’s Blackwell clusters.
This reality has triggered a race for the only emission-free source capable of constant output: Next-Generation Nuclear.
We are witnessing a series of historic transactions confirming this pivot:
Microsoft signed a 20-year deal to restart the Three Mile Island nuclear plant.
Amazon purchased a data center campus directly connected to the Susquehanna nuclear station.
Google is backing Kairos Power to deploy small modular reactors (SMRs).
But the most significant development is happening off the radar, in a legacy industrial sector in Wyoming.
The Gridlock Problem
Why are these companies bypassing the public grid entirely? Because the public grid is broken.
Currently, the "interconnection queue"—the waiting list to connect new power projects to the US grid—is backed up by an average of five to seven years. A tech giant cannot wait until 2032 to train a model that needs to be released in 2026.
This has forced Big Tech to adopt a "behind-the-meter" strategy. By colocating data centers directly next to power plants (as seen in the Amazon-Talen Energy deal), they bypass the transmission queue entirely. This makes existing power sites with industrial zoning the most valuable real estate in the digital economy.
The Wyoming Blueprint
Why Wyoming? The logic is one of infrastructure arbitrage.
Legacy coal towns possess the three assets that AI data centers desperately need:
High-voltage transmission lines (already built).
Water access (for cooling).
Industrial zoning (permits are faster to secure).
Bill Gates’ TerraPower project in Kemmerer, Wyoming, is the proof of concept. By replacing a retiring coal facility with a next-generation reactor, they are converting "stranded assets" into the most valuable real estate in the AI supply chain.
The Investment Implications
The market has largely priced in the growth of chipmakers. It has not yet priced in the utilities and specialized firms that will fuel them.
If the "AI Energy" thesis holds, capital will flow aggressively from overbought tech stocks into the suppliers of baseload power. This is not a speculative bubble; it is a supply-demand imbalance that will take a decade to resolve.
The winners of the second phase of the AI boom will not be the companies making the chips. It will be the companies keeping the lights on.
How would you rate today's investigation on Wyoming?
Written by Deniss Slinkins
Global Financial Journal
