Do data centers really increase electricity prices?

Start with how a utility actually sets rates. Take everything it costs to serve every customer in the territory, divide by the total kilowatt-hours sold, and that ratio drives the price. Now add a data center. It costs something to connect and serve, which pushes the top of that fraction up, but it also buys an enormous volume of electricity, which pushes the bottom up. If the bottom grows faster than the top, everyone’s rate goes down. Five years ago that held in a lot of territories, because many grids still had spare capacity sitting around and a new customer could be served without building much of anything.

That headroom is now gone. Serving a large new load today usually means new generation, new transmission, substation upgrades, real money on the top of the fraction. On the arithmetic alone the local answer became genuinely uncertain. What rescues it is that rates are not set by arithmetic alone. For a mid-sized utility’s entire load, a few-hundred-megawatt data center might add 10% or more while gigawatt-scale projects might be in the range of 30% or more. That makes it an extraordinary customer, and utilities have figured out that they hold the cards. They now negotiate bespoke deals that make the data center cover its own costs and then some. So the local answer lands back on neutral-to-slightly-positive, and utilities have started going to regulators asking to lower rates because of large loads. Early evidence supports it: an EPRI study found no correlation between data center presence and higher prices, and a small benefit to customers once you control for confounding variables.

Then the picture inverts when you zoom out to the global level. Lubershane is precise about the attribution here: data centers are one factor among many, though the biggest one, behind a surge in electricity demand fast enough to count as a genuine demand shock. Every link in the supply chain is bottlenecked and prices show it, with conductor roughly doubled, transformers more than doubled, switchgear doubled and gas plants two to three times more expensive. Renewables, which spent a decade pushing prices down, now cost more to interconnect while their tax credits wind down.

The rest of the pressure has nothing to do with AI at all. Tariffs, LNG exports raising domestic gas prices, general inflation and a rising cost of capital for an extraordinarily capital-intensive system all push the same direction and are largely orthogonal to anything data centers are doing. And then the detail that makes the whole thing bite hardest: roughly two-thirds of what US utilities spend is not about growth in any form. It is maintenance and hardening, replacing worn-out equipment, storm and wildfire work, and every dollar of it would be spent if the data center boom had never happened. That non-negotiable two-thirds now costs more too, because the supply chain squeeze raises the price of the same transformer whether it is going into a new data center or replacing a broken one. No local tariff negotiation touches any of this.

Which produces a genuinely awkward conclusion. Because the local effect is good and the systemic effect is bad, the rational thing for any individual community is to want as many data centers as possible in their territory. It is a prisoner’s dilemma: everyone competing for the local benefit while collectively driving the systemic cost.