Back in June, I wrote a column called “Folks, This Ain’t About the Data Center Building.”
The argument was pretty simple.
Everybody was staring at the big windowless building going up outside town.
I thought people were looking at the wrong thing.
The real story was everything required to make that building work.
Power plants.
Transmission lines.
Substations.
Natural-gas pipelines.
Fiber.
Battery farms.
Water systems.
Semiconductor plants.
Land.
Labor.
And enormous amounts of money.
I wrote that the United States wasn’t merely building data centers.
We were building an AI infrastructure network, something that might eventually be compared with railroads, rural electrification, interstate highways and telecommunications.
That seemed like a fairly big claim in June.
Three and a half months later, I’m beginning to think I was still looking too closely.
Because data centers may not be the story.
Electricity is.
First, $10.3 trillion
The Wall Street Journal recently put a number on the AI infrastructure boom.
Columbia University economist Stijn Van Nieuwerburgh estimates the United States could spend $10.3 trillion on data centers and related AI infrastructure between 2025 and 2032.
Ten-point-three trillion dollars.
That’s an average of roughly 3.6 percent of the American economy every year over the period.
Suddenly, those historical comparisons don’t sound quite so extravagant.
Railroads.
Electrification.
Highways.
Telecommunications.
Those weren’t merely new technologies.
They were construction projects.
They required enormous amounts of capital, labor and raw materials. They changed where people lived, where businesses located and how the country worked.
AI is beginning to look remarkably physical for something we keep calling “the cloud.”
But now comes another number.
And this one may be even more important.
Eighteen percent
Tim McDonnell, Semafor’s climate and energy editor, came away from New York Climate Week arguing that perhaps we should talk less about data centers.
Not because their electricity consumption has been exaggerated.
Quite the opposite.
What’s being built now is only a fraction of what’s in the pipeline.
Connor Teskey, CEO of Brookfield Asset Management, told McDonnell that if there is one market narrative most divorced from what Brookfield is seeing on the ground, it is the idea that demand for AI infrastructure and the energy supporting it is somehow fragile.
But then McDonnell pointed to an S&P Global estimate that puts the whole debate into perspective.
Between now and 2030, data centers are expected to account for only 18 percent of new global electricity demand.
Eighteen percent.
Which means we’ve spent an extraordinary amount of time arguing about the most conspicuous new customer while the other 82 percent was quietly standing in line.
Air conditioning.
Manufacturing.
Electric vehicles.
Electrified buildings.
Industry.
Homes.
Factories.
All of them want more electricity.
And suddenly the story becomes much bigger than Anthropic.
Data centers are the stress test
Maybe that’s the real significance of the data-center boom.
Data centers arrived first.
And they arrived fast.
A factory takes years to plan.
Millions of electric vehicles appear gradually.
Air-conditioning demand grows house by house and shopping center by shopping center.
A hyperscale data center can show up at a utility’s door asking for hundreds of megawatts.
Sometimes a gigawatt.
Sometimes several.
Right now.
That gets people’s attention.
I’ve been following projects measured at scales that would have sounded preposterous only a few years ago.
Three hundred megawatts.
A gigawatt.
Eleven gigawatts.
At some point you stop talking about plugging in computers and start talking about building an electrical system.
That’s why the industry has adopted one of my favorite phrases from this whole strange boom:
Speed to power.
Silicon Valley used to worry about speed to market.
Now some of the richest technology companies on Earth are asking:
How quickly can you build me a substation?
The cloud discovered the turbine
And when the grid can’t provide electricity fast enough, companies are beginning to provide it themselves.
We’ve already seen proposals involving behind-the-meter natural-gas generation.
That means putting generating capacity at or near the data center instead of waiting years for utilities to build enough generation and transmission.
From the developer’s standpoint, the attraction isn’t difficult to understand.
The AI chips are sitting there.
The building is ready.
Billions of dollars have already been committed.
The utility says the necessary grid connection may take years.
Somebody calls the turbine salesman.
Problem solved.
Except it isn’t.
The electricity problem has merely moved behind the fence.
McDonnell makes the larger point: if America cannot build generation and transmission quickly enough, the pressure for rapid economic growth will push companies toward solutions that may be more expensive, less efficient or more carbon-intensive than what we would build if the electrical system could respond faster.
Data centers are showing us that problem now.
The rest of the economy may show it to us later.
We’ve already seen the warning lights
Look at some of the stories I’ve been following.
In California’s Imperial Valley, a proposed data center wasn’t merely a 950,000-square-foot computer building.
It called for 330 megawatts of electrical capacity.
An 862-megawatt-hour battery system.
Four 500,000-gallon water tanks.
And 132 natural-gas generators.
It was seeking roughly 880 acre-feet of Colorado River water annually — about 287 million gallons.
I wrote then:
“At some point you stop describing a computer facility and start describing an industrial power plant with servers attached.”
That remains true.
But perhaps even that description was too narrow.
What we were really seeing was what happens when a huge new electrical customer tries to attach itself to an infrastructure system that wasn’t designed for growth at this speed.
Twenty-five gigawatts in six months
Then came another number.
JLL reported that North America’s data-center market absorbed 25 gigawatts of capacity during the first half of 2026.
Another 66 gigawatts were under construction.
About 95 percent of that construction was already committed.
Vacancy was around one percent.
I wrote about that in “AI May Live in the Cloud; Its Power Bill Doesn’t.”
At the time, I saw those numbers primarily as evidence of the extraordinary scale of the AI boom.
They still are.
But put them beside that S&P Global estimate and they begin telling another story.
If all this construction represents only part of the coming increase in electricity demand, what exactly are we preparing for?
An old problem wearing a new shirt
America knows how to build electrical infrastructure.
We’ve done it before.
Lots of it.
We electrified cities.
Then farms.
We built giant hydroelectric dams.
Coal plants.
Nuclear plants.
Natural-gas plants.
Thousands of miles of high-voltage transmission.
Substations.
Distribution systems.
An electrical network so dependable that most Americans rarely think about it.
Flip the switch.
The light comes on.
That may be one of the greatest technological achievements in American history precisely because we stopped noticing it.
But much of that system was built for a different age.
Now we’re asking it to accommodate artificial intelligence, advanced manufacturing, electric transportation, electrified buildings and growing cooling demand — while also replacing aging generating plants and transmission equipment.
And we’re trying to do it quickly.
That isn’t a data-center problem.
That’s an infrastructure problem.
Follow the money, but follow the megawatts too
The $10.3 trillion AI investment estimate matters because capital on that scale changes things.
It competes for workers.
Equipment.
Land.
Financing.
Turbines.
Transformers.
Transmission capacity.
And electricity.
But the 18 percent number may tell us something even more consequential.
Suppose AI investment slows.
Suppose some proposed data centers never get built.
Suppose investors discover that a few of those projected returns existed primarily in PowerPoint presentations.
We’ve seen that movie before.
The dot-com bubble burst.
The Internet didn’t disappear.
Companies failed.
The fiber remained.
Something similar could happen with AI.
There can be an AI investment bubble and a genuine need for electrical infrastructure at the same time.
Especially if data centers aren’t responsible for anything close to all the coming demand.
The grid may need enormous investment even if Wall Street eventually decides it built too many server farms.
And then there are the neighbors
There is another lesson from the data-center fight that shouldn’t be lost when we broaden the story.
Communities matter.
We’ve watched residents discover enormous projects after developers had already spent months assembling land.
We’ve seen nondisclosure agreements.
We’ve seen fights over water.
Electric rates.
Tax abatements.
Gas generators.
Noise.
Transmission lines.
And who pays for new infrastructure?
Those fights are sometimes portrayed as people standing in the way of progress.
That’s too easy.
People tend to become suspicious when somebody proposes spending several billion dollars next door and won’t tell them much about it.
McDonnell makes the same point about the larger power buildout: developers and their customers are going to need much greater transparency and accountability if they expect communities to accept the infrastructure that is coming.
That applies whether the customer is Meta or a factory.
Because if the next decade really requires a historic expansion of America’s electrical system, a great deal of it will have to be built somewhere.
Somebody will live near it.
Maybe we were looking at the locomotive
Back in June, I ended “Folks, This Ain’t About the Data Center Building” with a railroad analogy.
“Because folks, whether we like it or not, the railroad is already being laid.”
I still like that line.
But now I think we were staring at the locomotive.
AI is enormous.
The data centers are enormous.
The investment is enormous.
And the power demand is enormous.
But they’re revealing something even larger.
America — and much of the world — may be entering another great age of electrification.
Not because of one technology.
Because of everything.
All arriving at the same electrical system and asking for more.
The data centers simply got to the counter first.
So perhaps we should thank them for one thing.
They’ve shown us the problem early.
For months I’ve been saying:
Folks, this ain’t about the data center building.
Turns out it isn’t entirely about AI either.
It’s about electricity.
And we’re going to need a hell of a lot more of it.