What big data centers mean for your electric bill

Owen Quinlan is head of data at Arbor, a free service that helps consumers find lower rates on their electricity. That kind of offer may be more enticing than ever as AI systems—and all the data centers those AI systems require—are poised to suck up a lot of energy in the neighborhood.

OpenAI CEO Sam Altman, in a June 2025 blog post, wrote that one ChatGPT query equals approximately 0.34 watt-hours, “about what an oven would use in a little over one second.”

An April 2025 report from the International Energy Agency revealed that data centers accounted for around 1.5% of the world’s electricity consumption in 2024, and the IEA expects that number to double by 2030, consuming more electricity than the “production of aluminium, steel, cement, chemicals, and all other energy-intensive goods combined.”

An AI-focused data center holds a 100-megawatt capacity, “consuming as much electricity annually as 100,000 households,” the report said.

“AI is not going to go away. Building new data centers isn’t going to go away,” Quinlan told IT Brew. “People want it. We shouldn’t hold out hope that demand is going to temper.”

Quinlan spoke with us about the power demands of AI, and what’s limiting our supply.

Responses below have been edited for length and clarity.

How much of a data center’s cost goes to the Big Tech company, and how much gets passed along to consumers? Is there a loser?

Consumers are generally the loser…Generally speaking, large industrial has the lowest rate on a per-unit basis because they’re a large consumer, and it’s pretty guaranteed that they’re going to use a lot of energy. They generally have the lowest rate, but they are now a net-new consumer, and they’re just paying their regular price. So, that gets socialized out across all consumers, residential, consumer, industrial, the whole population. What happens, though, is that these capacity prices that forecast how much additional load is going to show up…hit everybody pretty much equally.

Do we have the energy infrastructure to support the rising demand from AI?

Almost definitely not. The grid that we have in this country is poles and wires that have been built over the last 120 years. A lot of them are fifty, sixty years old...Our transmission lines have been sort of lagging behind [the] new generation…For various red-tape reasons, it’s really hard to build a new transmission, which means it’s really hard to move electricity from where it’s being generated to where you want to consume it. As these new data centers pop up, they usually pop up near existing transmission. We just add additional strain to that [existing] underwater transmission system that we have.

What are the alternative energy options that can save the day here?

I’m an “all of the above” advocate. With the way that the demand is looking, we just need to build more electricity generation of all types. One that has gotten a lot of play recently, and is starting to pop up in meaningful ways in California and Texas, is batteries: large-scale energy storage. The classic challenge with wind and solar is that they’re intermittent. You can only generate it when the sun shines; if you need power and the sun’s not out, you’re out of luck. But that gets tempered with large-scale storage...That is a big one to focus on, because the time to develop it is short. Nuclear is great. We should build more nuclear plants. But it takes a decade, at least, from, “I want to build a nuclear plant” to, “I have an operational nuclear plant.”

How has the megabill impacted the “all of the above” option?

What happened with the recent bill is that the subsidies and tax credits available to people building solar, and less so wind, are being phased out over the next two years…In 2027, we’re going to see many fewer solar plants start to come online than what had happened under a different tax-credit regime. And that’s bad, because of the timeframe associated with all this and the ability to roll these things out quickly and meet the rise in demand.