Are Data Centers Making Electricity More Expensive? Not Next Door, But Across the Country

A power meter on the wall of a small-town house at night, with transmission lines stretching toward a brightly lit data center campus still under construction on the horizon

A single leaf over the eye hides Mount Tai; two beans in the ears block out the thunder. —— Heguanzi, “Tian Ze” (Warring States period; translated by the author)

What this episode is about

In the September 10, 2026 episode of Catalyst, host Shayle Kann brings back his colleague Andy Lubershane, head of research at Energy Impact Partners, to take on a question Andy’s own mom asked him: are data centers making my electricity bill go up? Andy’s answer is “no, and yes.” Both hold at once, as long as you split the books in two: one ledger for the utility that serves your home, and one for the electricity system as a whole.

Key takeaways

1. An electricity rate is a fraction. The numerator is what it costs to serve all customers; the denominator is the total kilowatt-hours sold. A data center grows both. Five years ago, when grids had headroom, hooking up a data center rarely required new power plants, so it added more to the denominator than the numerator and spread costs thinner for everyone.

$Three fractions sit side by side with cost on top and kilowatt-hours on the bottom: the original 10 divided by 10 equals 1.0; when the grid has spare capacity, connecting a data center adds many more kilowatt-hours but only a little more cost, so the price per kWh falls to 0.8; when the grid has no spare capacity, cost jumps sharply and the price per kWh rises to 1.2.$

2. Once the headroom is gone, it comes down to the negotiating table. Most grids now have no spare capacity. A new large load means new generation, substation upgrades, new transmission — the numerator jumps too. Whether rates rise or fall then depends on the special tariff the utility negotiates. A data center of a few hundred megawatts can add 10–30% to a mid-size utility’s total load. Contracts for customers like that are negotiated one by one, and utilities ask them to pay their full share plus a buffer, or bring their own capacity.

3. Three phases. Shayle breaks the last few years into three stages. In the cloud-and-Bitcoin era, grids had headroom and states offered tax breaks to attract data centers. In the early AI boom, utilities hadn’t yet learned what they could charge, and the local evidence was mixed. Now, in phase three, utilities know they hold the cards; announcing a new data center interconnection alongside a promise that rates won’t rise — and will likely fall — has become standard. A few utilities have already asked regulators for rate decreases, citing large loads in their territory.

Three seesaws are lined up in order, with data centers on the left end, utilities on the right end, and bargaining chips stacked on the heavier side: in phase one the data center end sinks, in phase two the two ends are level, and in phase three the utility end sinks, showing that bargaining power has moved from data centers to utilities.

4. The data backs “not locally.” A recent EPRI study compared utility territories across the U.S. Without controlling for anything, there was no visible correlation between data center load and higher prices. With controls, data centers were on average lowering customer bills slightly — something like 6%.

5. The national ledger is a different story. Data centers are the single largest driver of today’s electricity demand growth, and it is moving so fast that it amounts to a demand shock within the sector. The supply chain can’t keep up: conductor costs have doubled, transformers are more than 2x, switchgear 2x, gas plants 2–3x; Shayle recently saw a combined-cycle gas plant quoted at $3,600 per kilowatt. Worse, about two-thirds of U.S. utility spending goes to maintaining and hardening the existing grid — spending that would happen with or without data centers — and that two-thirds is now getting more expensive too. No local tariff can shield customers from this layer.

A bar chart uses the original price as a 1x baseline: conductors and switchgear have both risen to 2x, transformers are above 2x, and gas power plants fall in the 2x to 3x range, showing that grid equipment has become more expensive across the board.

6. There are other forces too. Renewable tax credits expire over the next four or five years while interconnection costs climb; tariffs and LNG exports push up domestic gas prices; and higher interest rates make the grid — a once-in-a-generation capital project — more expensive to build. Andy says rates worry him most: this investment would have been far better made ten years ago, when money was cheap.

7. Lower bills don’t buy back trust. In a Gallup poll, the top reason people oppose data centers was water use; energy use and higher electricity bills came second and third. When Shayle asked friends and family “what if the utility guaranteed your bill would go down?”, he hit a deeper layer: they don’t trust the data center companies, the utilities, or the politicians. A data center can keep your rates lower than they’d be without it while overall rates still rise — and since nobody sees the counterfactual, nobody gets the credit.

Further thoughts

When I read “A makes B more expensive,” I first ask: compared to what?

When I see a headline like “data centers are driving up power prices” or “this policy raised home prices,” my instinct is to pick a side: believe it or don’t. This episode showed me that sentences like these hide two different baselines.

The first is compared to the past: how much more is this year’s bill than last year’s. The second is compared to a world without it: same day, same place, minus that data center — what would the bill be? The local ledger answers the second question, and the answer is that you pay a bit less. The national ledger answers a different “without it”: without this wave of AI power demand, equipment wouldn’t be this scarce and everyone’s rates would be lower. People arguing about this are often each holding a different ledger.

Three points sit on a price axis: last year at 5, this year's actual at 6, and this year without data centers at 7 a dashed hollow point; the arrow from last year to this year points up, while the arrow from "without it" to this year's actual points down, so the same bill counts as a rise against one baseline and a drop against the other.

Stack the two ledgers and a prisoner’s dilemma appears. Shayle followed the logic out loud: so everyone should want as many data centers in their territory as possible? Andy said that in this specific context, yes. Every territory comes out ahead by landing one; when every territory does it, equipment costs climb for all. Individual best moves don’t add up to the collective best outcome.

Four districts sit in the top row, each box holding a downward arrow for a falling local electricity rate; four lines flow down into one shared equipment and labor market, where an upward arrow shows that total cost rises for everyone.

I think the same split works for reading earnings. A company whose gross margin fell two quarters running looks worse against its own past; against peers all absorbing the same input-cost spike, it may have held up best. Whether that decline is noise or structure depends on getting the baseline right first. I still tend to tense up the moment a number goes down; this episode put “compared to what?” back at step one for me.

Demand exploded — where does the money stop?

When I heard that every piece of equipment had gotten pricier, my first thought was: “So the power companies must be making a fortune?” The episode’s answer is no, and the reasoning is worth unpacking.

The chain runs like this: a demand shock → shortages in conductor, transformers, switchgear, turbines and labor → higher costs for new builds and maintenance → all of it spread into every kilowatt-hour. Utilities sit at the end of that chain. They are regulated; rates are approved by regulators, so rising costs get passed through rather than turned into profit — and passing them to households runs straight into the wall of public opinion. “Big demand” doesn’t mean “everyone in the chain gets rich.” The money settles in the layers where supply can’t keep up and sellers can set their own price.

A demand shock pushes from the left through three links: equipment and labor, utilities, and household bills; below them, with zero as the baseline, the equipment and labor column has a tall bar pointing up to show it profits, the utility column is almost flat because the money only passes through, and the household column has a bar pointing down to show it pays the bill.

Which layer stays tight longest? Andy’s view is that “the cure for high prices is high prices”: expensive transformers draw in new manufacturers and alternatives like solid-state transformers; price calls supply forth. What worries him is labor. Finding and training a master electrician takes years, and no price can speed that up. Shayle countered that electricians are a small slice of the cost of power. Andy’s reply stuck with me: it’s like energy for data centers — it’s everything and it’s nothing. A small share of cost, but without someone to connect the line, the new substation just sits there.

That changed the yardstick I use for bottlenecks. I used to judge an input by its share of total cost: the bigger the share, the more it mattered. This episode added two other questions: does everything stop without it, and the smaller its share of the buyer’s costs, the less the buyer cares about its price. Electricity is only 5–10% of a data center’s total cost. So when Vivek Ramaswamy, running for governor of Ohio, proposed that data centers should mean free electricity for local residents, Andy didn’t laugh it off; he asked what happens if data centers pay double or triple. Shayle went further: he’d love to see a small utility issue an open call — here’s a site, fast interconnection, and in exchange you pay a rate high enough to cut every customer’s bill by 25% — and see who takes it.

It also matters who ends up paying. Nat Bullard’s annual deck has a chart showing that the share of income Americans spend on electricity has been stable for about seventy years; since 2010 it has sat around 1–1.5% of personal income. That average is low, but for low-income and fixed-income households, electricity can be 5–10% of their budget or more. A small rate increase is a rounding error for the average family and grocery money for them. That’s why the political pressure is larger than the averages suggest.

References

Catalyst with Shayle Kann, “Do data centers actually increase electricity prices? No…and yes,” Latitude Media, 2026-09-10
Andy Lubershane’s analysis of data centers and electricity prices at Energy Impact Partners (the piece mentioned at the start of the episode)
Andy Lubershane, “For AI, energy is everything and energy is nothing”
EPRI research on data center load and local electricity prices (2026)
Edison Electric Institute statistics on utility spending
Gallup polling on public attitudes toward data centers
Nat Bullard’s annual deck, chart on electricity’s share of personal income

One thing to take with you

Whether something is good or bad for you depends on two ledgers: compared to before, and compared to a world without it. Most complaints only tally the first.

One thing I’ve tried: pick something that’s been bugging you lately — a rent increase, a new recurring meeting at work, a new teammate. Write two lines on a piece of paper: “Compared to a year ago, it makes me…” and “Compared to today without it, it makes me…”. If I can’t fill in the second line, I don’t yet know whether I’m upset about the thing itself or about the whole environment shifting around it.

Raw Markdown

This article is an educational discussion of investment method. It is not advice to buy or sell any individual security, offers no target prices, and does not analyze any current holding. Investing carries risk; make your own decisions or consult a qualified professional.