How do we make electricity cheap again?

The views expressed here are my own and do not reflect the views of Powerhouse Innovation.

The path to decarbonization is through electrification, and yet power prices seem to have gone in the wrong direction lately. Good people have helped explain the nuances (including the most annoying: “well actually, adjusted for inflation, prices are about flat”). But if we really care about decarbonization, it seems we should ask: how do we Make Electricity Cheap Again?¹ I think the simple and yet counterintuitive answer is that we must start consuming a lot more of it. If you’re an abundance enthusiast, an environmentalist, or even just a nostalgiast, there’s something in here for you. 

Electricity : Action :: Money : Value

Electricity is deservedly having a moment. Electricity is the highest quality form of energy because it’s easily converted into every other form of energy (sometimes at well over 100% efficiency). You might call this exergy. Electricity can do lots cool things, like make light come out of crystals, draw heat from winter air, or make thinking brains out of sand. So even if you don’t care about climate change, you should like for electricity to be a bit cheaper.

We need to consume a lot more electricity

But we do care about climate change and most decarbonization models have us more than doubling total electricity consumption (largely from electric vehicles, heat pumps, and industrial electrification, but that was before LLMs were a thing). The U.S. consumed over 4,000 TWh in 2025, peaking at 759 GW. So imagine adding another 6,000 TWh, pushing peak load as high as 2,000 GW. That would dwarf even the data center boom that we’re having today (it depends on who you ask, but utilities think it’s 90 GW of peak load by 2030, others estimate 750 TWh per year). 

Hourly electricity demand in New England with and without electrification for deep decarbonization. From the supporting information in Nestor Sepulveda’s legendary paper, The design space for long-duration energy storage in decarbonized power systems.

How did electricity get cheap the first time around?

For over 100 years we enjoyed decades of load growth, a growing grid, and dramatically falling prices. And then we stopped. When the legendary Sam Insull got into the electricity business at the turn of the 20th century, electricity cost $0.16/kWh (nominal, ~$6/kWh in real terms). Over the next 70 years, prices dropped to $0.02/kWh as annual consumption shot above 7,500 kWh per person. But the 1970’s energy crisis marked the first decade that electricity prices went up (both in real and nominal terms), and electricity growth slowed down. Since then, nominal prices have continued to creep up, especially lately, but not meaningfully faster than the broader inflation index. 

One of Insull’s key insights was that different customers tend to use electricity at different times.² The trolleys ran more during rush-hour, the factories ran more during the day, and the lights came on at night. So if he put all these customers together, he could drastically increase the amount of electricity sold without having to increase capacity by nearly as much. Improving this diversity factor took their plant’s load factor from 5.5% in the 1890s to over 50% in 1910. This created a positive flywheel, where the bigger the grid got, the cheaper it was to serve everyone, which made it more appealing to new customers, and grew the grid even further. This is when rate-of-return regulation was invented, and it worked pretty well as long as demand was growing. 

We don’t use the grid like we used to

The problem is that return on equity and rate of return don’t work as well when demand stops growing. Even though demand slowed down, utilities continued building, and power prices went up. Depending on who you ask, capacity utilization has fallen from about 95% to 70%, or even as low as half. Load factor for El Paso Electric has been trending down since 2000 (source, via Halcyon), now around 45%. This is not great! Running a utility is more like manufacturing than most commodities: you’ve got huge overhead that is very unhappy if the factory sits idle.

Peanut butter the load

The solution is the same as what Sam Insull did, but maybe without the suicidal leverage and bankruptcy. Our greatest minds are now working on a nationwide treasure hunt to find which pockets of the grid have spare deliverable capacity. This is great, and should be encouraged. We must peanut butter the load, to fill those gaps in time and place (be they large, small, or tiny). Utilities want this load! To the extent that they will fight for the privilege of supplying the canonical flexible load: rock crushers.

Even if we do more than just peanut butter, I think AI overbuild is good for two reasons: flexibility and fat pipes. Having a bunch of excess capacity that can shift compute across the country would be great. And stranded grid assets might even be good. If a utility foolishly builds an oversized substation, then that can become a cheap interconnect for something that’s actually useful like an EV charging depot or a manufacturing plant (I’m just kidding, I don’t think AI is useless).

The only way out is through

You can’t do one without the other: we can’t MECA if we don’t do load growth, and we can’t do load growth if we don’t MECA (or we’ll deservedly have a revolt on our hands). Obviously, power prices also came down because the power plants and infrastructure got cheaper, and we want to stimulate more of that through learning rates on new technologies. But that doesn’t work without load growth.

What could go wrong? The thing I worry about most is that all this demand for transformers and gas turbines doubles costs for utilities, which gets passed on to consumers. Data centers are effectively bidding up the price for gas turbines, and if they’re off grid then not only are they overpriced, they’re not even shared with consumers. Better rate design can make sure that large loads cover these excess costs, but that takes good policy.

1. Saul Griffith and Michael Liebreich are also on the MECA train. They have a lot of great ideas, but this is mainly just about one idea. Let’s make MECA a thing! ↩︎
2. This phenomenon is not restricted to power. Much like how utilities like uncorrelated loads, investors love to have diversified portfolios of uncorrelated assets. Anthropic lets you pool usage to maximize utilization. I wonder if data centers do this too: LLMs during the day, binge Netflix at night? ↩︎