Electricity Demand is going to Skyrocket

It's not 2% - 3%/year, think 20%/year for the next decade

Yes, yes, yes datacenters are a very small part of our electrical use today. And so significant growth there is not large compared to the entire grid. And up to a certain point, that’s valid.

But people are not understanding the size and scope of the required increase for AI. It’s so large that it will become a significant part of our electrical usage. And with that growth, as it becomes larger, it’s growth becomes larger.

At the same time we’re moving the rest of our world to electricity. EVs are becoming an ever larger part of the vehicles on the road. Electric heat pumps are replacing the various forms of furnaces we use. At a certain point it’s no longer profitable to have a gas station or to deliver gas or heating oil to homes.

Think I’m overstating it? There is presently 2.6TW of grid connection backlog. The grid presently produces 1.3TW of power. The backlog today is 200% of our present generation capacity. While some of that backlog may go away, there’s also people not bothering to request additional power because the line is so long.

My estimate of 20%/year is probably understating it.

The article A Primer on AI Data Centers illustrates the size and scope of the need/demand today. This is not estimates for the future, this is datacenters they would build tomorrow if power was available.

AI is not some “maybe someday” it will be of use and pay off. It’s in widespread use today. This is akin to when we built out the electrical grid in 1900. The scope and price of that buildout was unimaginable. And yet, over the first 10 years, electricity was brought to pretty much every company, factory, and home in urban areas.¹

This A.I. is being used, across industries. 25% of the Y Combinator Winter startups wrote 95% of their code with AI. Most research, including for my blog here, is now performed by AI as it can deliver a more thorough and accurate report than a human. The list goes on.

To a large degree, effectiveness and efficiency at work is determined by how well the employees use AI. And how often. Which means The Gigawatt Data Center Campus is Coming. Not one, but many of these. Wrap your head around the fact that five of these are the same as the power New York City requires. And just as New York is the city that never sleeps, these data centers will be running 24/7.

What does this mean for power? First off, we need to look at AI as electricity. It’s something we’re constantly using and are dependent on it always being available. It’s not something that can be stored or saved like water, it’s something that when we ask it needs to be available to respond.

Second, it’s rapidly becoming as important and central to our lives, as well as to companies, as electricity and the internet is. It won’t be acceptable to tell individuals and companies in a couple of years that they get limited AI compute because we haven’t built enough power. Tell them that and the voters will insist that we turn all the coal plants back on.

And same for the Big Tech companies. They know their future depends on their being on the leading edge with AI. They know that means power. And while they’d prefer that power be green, they’ll take whatever they can get. And they have the money and the determination to get that power now.

We need to be on a crash course, starting yesterday, building out Giga-Watts of green power yesterday.

1. Rural areas had to wait for FDR and the REA.