The Data Center Pays Less Per Kilowatt Than You Do. I Went Looking For Why.

Turns out we're not buying the same thing

A kilowatt-hour is a kilowatt-hour. That was my assumption walking in, and it falls apart about ten minutes into any utility rate filing.

The electricity itself is a small slice of what you pay. The rest is the machine that carries it to your kitchen — the substation, the wire down your street, the transformer humming on the pole out front that serves maybe six houses, the meter, the guy who comes out when a squirrel takes the whole block down. A big data center skips almost all of that. It connects way upstream at transmission voltage and builds its own substation on its own dirt. Anybody who has watched a package sit at the sorting facility for a day and then take three more days to cover the last eleven miles to their porch already knows which part of delivery is expensive.

Then there's when you use it. Regulators score every customer on load factor, which is just how flat your usage is across the day. A data center draws nearly the same at four in the morning as it does at four in the afternoon. Virginia wrote that straight into its new rate class for giant loads: twenty-five megawatts or more on one site, running at a load factor of at least 75%. My house is the opposite animal. Dark all day, then the entire neighborhood walks in and hits the AC and the oven inside the same forty minutes, in July.

The grid has to be built for that forty minutes. It's a gym in January. Management buys enough treadmills for the first week of the year, those treadmills sit idle until next January, and every member pays for them all twelve months. The member who shows up at 2 a.m. every single night is using machines nobody else wanted. Cost-of-service studies say that member is cheaper to serve, and on the math, they're right.

So who actually picks the number?

I assumed the rate came out of a handshake. It mostly doesn't. Before rates change, a utility files a cost-of-service study that splits every dollar of its revenue requirement across customer classes — residential, small commercial, large industrial — based on how many customers each class has, how much energy it pulls, and how hard it leans on the system at peak. Classes whose revenue lands below what they cost to serve get pushed up in the next case. It's a public proceeding. The finished rate is a published tariff sheet with a number on it, and you can go read the one that applies to the building down the road.

Usually. Here's where my tidy explanation cracked. Harvard's Electricity Law Initiative went through close to fifty state commission proceedings and found that a lot of very large customers don't take a published tariff at all — they sign a special contract with the utility, filed under confidential treatment, terms sealed, and commissions have approved a bunch of them in short orders. Their phrase for it is secret contracts between utilities and data centers. You can't check a rate you're not allowed to see.

And there's a motive in the plumbing that nobody hides. A regulated utility earns a return on the capital it builds. Bigger load justifies more poles, more substations, more generation, which is more capital, which is more return. That's not a conspiracy, it's the business model working exactly as designed — but it means "we need to build this for the new customer" is an argument the utility is structurally happy to make.

The rate was never where the money was

Here's the turn that reorganized the whole thing for me. Your bill can be climbing hard while that data center's per-kilowatt rate stays perfectly defensible, because the money isn't moving through the rate. It's moving through the capacity market.

In the PJM region — the grid stretching across thirteen states and D.C. — utilities buy commitments from generators to stand by and be available. That standby price has gone up more than tenfold in about two years. PJM's own independent market monitor, whose entire job is calling out what the market is doing, attributed most of one auction's increase to data center load and put the resulting cost at $9.3 billion, recovered from customers across the region in higher rates. Through the middle of 2026 the monitor pegged data center load at roughly nine percent of the wholesale price of power in that footprint, and that figure doesn't even include the transmission build.

That cost doesn't land on a data center line item. It lands in everybody's supply charge, because it's the price of electricity itself going up when demand outruns what's plugged in. So the comment-section complaint and the actual problem are two different stories that keep getting mashed into one. The rate sheet is roughly fair. The bill is still going up.

And then there's the version that keeps regulators awake: a utility builds the substation and signs for the generation, the AI buildout cools off, the project gets canceled, and the concrete is already poured. Somebody pays for that hardware either way. Somebody is always a retiree three towns over who never heard of the company that walked.

What I couldn't settle, and where I'm going next

I want to be square about the part I couldn't close. Researchers at EPRI and Watershed ran the numbers on 2015 through 2024 and found the opposite of what I expected: doubling a state's data center capacity lined up with residential rates going down about three and a half percent, because a steady round-the-clock customer spreads the grid's enormous fixed costs over more kilowatt-hours. Their own caveat is the whole ballgame — turbine backlogs, transformer shortages, permitting that won't move. When you can't build new supply fast enough, the effect flips. One study is looking at the decade before the crunch. The market monitor is looking at right now. From my kitchen table I can't referee that, and anybody who tells you it's settled is selling something.

What I can watch is the rulemaking, and it's moving fast. More than twenty states now have a large-load tariff on the books. Virginia's takes effect in 2027 and it's blunt: a fourteen-year contract, a minimum monthly demand charge whether you draw the power or not, three years' notice before you can shrink, and one and a half million dollars per megawatt posted as collateral before anybody pours a slab. Oregon approved a framework where the biggest projects sign for up to thirty years and pay a surcharge of a penny per kilowatt-hour that goes back toward residential bill relief. Pennsylvania's commission built its own. That's regulators trying to make the promise cost something up front.

That's the rabbit hole I'm filming on the Byte Bungalow channel — the gas turbines with multi-year waiting lists, the substation going up beside somebody's subdivision. Come watch the videos and hit subscribe so you catch the next one. And if you can find the line on your own bill where capacity costs actually live, put it in the comments, because I've read mine four times and I still can't point at it.

Common questions

Keep reading