How the Lights Stay On

Chapter 11: The Grid That Has to Grow

In Loudoun County, Virginia, about forty miles west of Washington D.C., there is a stretch of suburban road that looks, at first glance, like an ordinary American office park. Big rectangular buildings sit behind landscaped berms. Parking lots are mostly empty. There are no signs identifying what is inside.

This is one of the most important places in the modern world.

The buildings are data centers. Inside them are millions of servers — the physical hardware that runs a substantial portion of the global internet. When you stream a movie, send an email, search for something, or ask an artificial intelligence a question, the request very likely passes through Loudoun County. The local industry trade group estimates that about 70 percent of the world’s internet traffic touches a data center somewhere in northern Virginia. By most measures, Loudoun County hosts more data center capacity than any single place on Earth.

In 2025, Loudoun County’s data centers drew about 4 gigawatts of continuous electricity — roughly the same amount of power as four large nuclear plants running flat out. The county’s total demand for data center power has grown by about 30 percent per year for the last several years. By 2030, the demand could be 12 gigawatts or higher.

To put this in perspective: the entire state of Virginia, a state of 8.6 million people, currently uses about 17 gigawatts at peak demand. The data centers in one county will, within a few years, demand more electricity than the rest of the state’s homes, businesses, and industries combined.

Somewhere in an electric utility’s office in northern Virginia, there is a planner trying to figure out where this electricity will come from. That planner has spent most of their career planning for an electric grid that did not need to grow.

Until very recently, no American electric planner had needed to plan for growth at all.

This may be the most surprising fact in this entire book.

From the year 2000 to the year 2020, total United States electricity demand was almost flat. It went up a little. It went down a little. Over the course of two decades, it grew by less than 5 percent in total.

Two things happened simultaneously to produce this strange equilibrium. The American economy kept growing. New houses kept getting built. New devices kept getting plugged in. Computing power exploded. By all the old logic of electricity demand, the country should have been using a lot more power year after year.

But it was not, because efficiency was getting better just as fast.

The shift from incandescent light bulbs to compact fluorescents and then to LEDs reduced the energy used for lighting by something like 80 percent over those twenty years. Refrigerators, washing machines, air conditioners, and televisions all became dramatically more efficient. Heating and cooling systems improved. Factories deployed better motors and better insulation. The country was doing more with the same kilowatt-hours, year after year, and the net result was that the grid did not need to grow.

This was, generally speaking, considered a triumph. It meant that economic growth no longer required ever-increasing energy use. It meant that climate concerns could be addressed without slowing the economy. It also meant that the electric utility industry, for two decades, mostly stopped building new power plants. The plants from the 1970s and 1980s were still there. They were enough.

For an entire generation of grid planners, this was the world. No new generation needed. Transmission upgrades only where things were aging out. Utility executives became experts at managing flat demand, optimizing existing plants, and squeezing efficiency. They did not become experts at building.

Then, sometime around 2022, the world changed.

The flat-demand era ended fast.

By 2023, every credible forecast was showing demand growing. By 2024, the forecasts were getting steeper every quarter. By 2025, electric utilities across the country were filing emergency requests with state regulators to build new generation, to upgrade transmission, to raise rates to fund the build-out.

A leading consulting firm, ICF, calculated in 2025 that US electricity demand would grow 25 percent by 2030, compared to 2023. Other forecasts put the growth higher. The International Energy Agency officially declared the world to have entered an “Age of Electricity,” with global demand growing faster than at any time since the postwar build-out of the 1950s and 1960s.

Three forces are driving the growth.

The biggest single driver, by most estimates, is data centers — and particularly the data centers built for artificial intelligence.

In 2024, US data centers consumed about 183 terawatt-hours of electricity — about 4 percent of total US electricity consumption. By 2030, depending on whose forecast you read, that number could rise to 9 percent, or 11 percent, or even higher.

The reason AI specifically demands so much electricity is that training and running the largest AI models requires enormous amounts of computation. A single AI training run for a state-of-the-art model can consume as much electricity as several thousand American homes use in a year. AI inference — actually running a trained model to answer your questions — uses less per query, but is happening billions of times per day across the world.

Globally, the International Energy Agency forecasts that data center electricity demand will grow from 415 terawatt-hours in 2024 to 945 terawatt-hours by 2030 — more than doubling in six years. The United States and China together will account for almost 80 percent of that growth.

All 18 chapters of How the Lights Stay On · Library · All courses