Why a Megawatt Is Harder to Find Than a Chip

On July 9, 2025, the Public Utilities Commission of Ohio issued an order directing AEP Ohio, the Ohio arm of American Electric Power, to create a separate set of rates for new data centers. Under the resulting tariff, large new data center customers with loads of 25 megawatts or more agree to pay for at least 85 percent of the capacity they reserve, whether or not they use it, for up to twelve years, including a ramp-up period. AEP Ohio had paused new data center service agreements in central Ohio while the case was decided. That order is a useful place to start, because it shows in a single document what the power problem looks like from the utility's side of the table.

A utility builds substations and transmission lines on the expectation that customers will use them for decades and pay for them over that time through rates. When a new customer asks for hundreds of megawatts, the utility faces two risks. The first is that it cannot build fast enough. The second is that it builds, the customer changes its plans, and everyone else on the system is left paying for equipment nobody needs. The Ohio minimum-demand requirement is aimed squarely at the second risk. It asks the data center, not the neighbors, to carry the cost of a forecast that turns out to be wrong.

The first risk, speed, is the one most builders feel. A new large load usually has to be studied by the utility and, in many regions, by the grid operator that runs the high-voltage system. In much of the eastern United States that operator is PJM Interconnection, which covers all or part of thirteen states and the District of Columbia. Studies check whether the existing wires and transformers can carry the new load without overloading anything, and what must be built if they cannot. Heavy equipment takes years to procure: the Department of Energy's July 2024 Large Power Transformer Resilience Report to Congress found lead times of around three years commonly quoted for large power transformers, and some as long as five. None of this can be shortened by spending more on servers.

This is why the phrase you will hear most often in the industry is time to power. A developer can buy land in months and put up a building shell within a year or two. If the utility says the substation serving the site will not be ready for several more years, the building waits. For an investor, the value of a site therefore depends less on its acreage than on the date a credible party has committed to energize it, and on what the customer has agreed to pay for that commitment.

The second consequence is geographic. Because capacity is local, the same company can face very different answers in two counties a few hours apart. A site next to a large substation with spare capacity may be energized quickly; a site in the most crowded market may wait much longer. Faraday's read is that this is one reason data center construction has spread into places that were not traditional hubs, and one reason regulators in those places have begun writing rules like Ohio's. Central Ohio, where AEP Ohio paused new agreements while the tariff case was decided, is an example of both.

There is a public-interest dimension here that is easy to overlook if you only look at the project. The cost of new grid equipment, and the price of generating capacity bought through market auctions, is shared among all customers of a utility or region. When large new loads arrive faster than supply, regulators and consumer advocates ask who should pay. Ohio's answer was to put more of the risk on the new large customer. Other states and utilities have been working through versions of the same question, and a reader should expect the answers to keep changing.

How does a newcomer judge a claim that a project has power? Start by separating three things that are often blurred. There is a request, which is an application to the utility. There is a study result, which describes what must be built and roughly when. And there is a signed agreement, which commits both sides to a schedule and a set of payments. Only the last of these is worth much, and even then the schedule depends on equipment arriving and construction going to plan.

Faraday's read is that power has become the pacing item for artificial intelligence infrastructure not because the country is running out of electricity in the aggregate, but because the specific equipment needed to deliver large blocks of it to specific places cannot be built as quickly as the demand is arriving. The Ohio order is one regulator's attempt to price that mismatch fairly.

With that constraint in view, the course moves inside the fence line, to follow electricity from the utility's wires to the server.