Last month I wrote about whether artificial intelligence could help fix the climate. This month I want to look at the same subject from the ground. Literally from the ground: the farm country and small towns where the AI boom is physically being built. From down there, the story looks less like software and more like the largest infrastructure rush in a generation.
The numbers coming out of the industry this autumn have detached from all precedent. Chipmakers and AI labs are announcing computing deals denominated not in dollars or chips but in gigawatts: ten gigawatts here, six there, as if they were utilities planning national fleets. A single gigawatt is roughly the output of a nuclear reactor. When a technology company casually commits to deploying "at least ten gigawatts" of computing, it is announcing that someone, somewhere, must conjure the equivalent of ten reactors' worth of electricity. I will admit my first reaction to those announcements was awe — engineers are trained on scale, and scale trains us back — so let me discipline it into the second reaction, which is a question: who is the someone? The announcements never dwell on it.
A Factory That Turns Electricity Into Arithmetic
A modern AI data center, minus the mystique, is a factory that converts electricity into arithmetic. Racks of specialized chips run hot, around the clock, training and serving models. The buildings themselves are unremarkable, windowless boxes of concrete and steel. What is remarkable is the duty cycle: a large AI campus can draw as much power as a mid-sized city, every hour of every day, concentrated on a plot of land that five years ago grew corn or soybeans.
This explains where the boom is landing. The industry needs three inputs on one site: cheap land, fiber-optic connectivity, and above all, available power. That combination is found not in Silicon Valley but in rural counties, places with strong transmission lines, agreeable zoning boards, and local governments hungry for tax revenue. The data center in the cornfield is not an anomaly. It is the pattern.
The Largest Load Step in Decades
For two decades, American electricity demand barely grew; efficiency gains soaked up whatever new uses appeared. Utilities planned accordingly, retiring old plants and building modestly. In control terms, the grid was tuned for a flat input. The AI boom has broken that assumption in the space of about two years. Utilities in data-center corridors are now fielding connection requests that exceed their entire existing load: more demand knocking at the door than they currently serve.
Something has to give, and three things are giving at once. Utilities are delaying the retirement of fossil plants that were scheduled to close, and dusting off plans for new gas turbines, the fastest reliable capacity we know how to build. Grid operators are warning about reliability margins. And in some regions, the costs of new infrastructure are beginning to seep into the electricity bills of ordinary households who have never used a chatbot. That last point deserves more attention than it gets. The boom's financing is corporate, but its grid is shared, and when a region's rates rise to serve its newest, largest customer, everyone pays retail for someone else's industrial revolution.
The Optimist's Case, Bias Disclosed
Readers of this column know I am, on balance, a technological optimist. I flag the bias so you can apply your own correction; I am going to make the case anyway. The AI industry is also the largest new buyer of clean energy on Earth. Tech companies sign enormous contracts for solar, wind, and battery storage; they are bankrolling the revival of shuttered nuclear plants and placing the first serious commercial bets on next-generation reactors and geothermal. Their demand is doing what decades of advocacy could not: making firm, clean power a product with impatient, deep-pocketed customers.
The machines are moving targets too. Each chip generation performs more calculations per watt; better cooling and smarter scheduling squeeze more work from the same megawatt. The industry's defenders argue, plausibly, that we are watching a messy construction phase, not a permanent trajectory.
So the ledger, as I keep it. The win: a wall of corporate money pointed at exactly the firm, clean generation the grid has needed for decades. The cost: in the near term, the boom is powered by whatever can be plugged in fastest, and what can be plugged in fastest is often gas. The catch: the win compounds on a ten-year schedule while the cost burns on a two-year one. Emissions deferred are not emissions avoided. A cleaner grid someday does not cancel a dirtier grid today.
Spec Sheet for an Honest Build-Out
Three requirements would make this build-out trustworthy, and none of them is exotic. First, transparency: companies announcing gigawatt-scale ambitions should disclose, with the same fanfare, where the power comes from and what it emits. Not annual averages laundered through certificates; hour by hour, grid by grid. Second, additionality: every large data center should be matched by genuinely new clean capacity in the same region, so the boom builds the grid it uses rather than borrowing everyone else's. Third, fairness: the costs of connection and expansion should land on the customer who caused them, not be socialized onto households through the boilerplate of rate cases.
All of it has precedent. The open question is whether an industry sprinting for competitive advantage will accept constraints it is not forced to accept, which is, historically, a question with a known answer. That is why the zoning boards, utility commissions, and county councils of farm country suddenly matter to the future of the climate. The AI revolution turns out to run through the most unglamorous rooms in American democracy.
The cloud, we keep relearning, is a place. This year, the place has a cornfield next door, a substation being expanded, and neighbors with questions. They are the right questions. The industry that answers to them honestly will deserve the future it is building.
