Everybody’s talking about how much money’s being spent on artificial intelligence.
And they’re right. It’s a ridiculous amount of money.
More than $1 trillion has already been spent since 2024 building the infrastructure for AI.
That spend spans computer chips, data centers, electricity, construction, and networking equipment.
And the numbers keep getting bigger.
At some point, when people start throwing around trillions of dollars, the numbers almost stop meaning anything.
But there’s another way to look at this.
What if all this spending that seems so crazy today isn’t evidence that we’re near the end?
What if we’re still near the beginning?
Every major technology revolution needs somebody to build the stuff first.
Think about the railroads.
Before you could put people and products on trains and move them across America, somebody had to build thousands and thousands of miles of track.
They needed steel. They needed bridges and tunnels. They needed train stations and locomotives. They had to buy land.
None of that was cheap.
During the railroad boom of the 1860s and early 1870s, roughly $2.5 billion was spent building American railroads.
That sounds like nothing today.
But the entire U.S. economy was only around $10 billion at the beginning of that boom.
In other words, the amount eventually spent building railroads was equal to roughly one-quarter the size of the entire U.S. economy at the start of the cycle.
That’s a lot of money for train tracks.
Then America did something similar as electricity spread across the country.
Power plants had to be built. Transmission lines had to go up. Factories needed new equipment.
The new technology was exciting.
But somebody had to pay for all the boring stuff that made it work.
Sound familiar?
Remember the Internet?
It’s easy to forget how much physical stuff had to be built for the internet.
Today I can pull my phone out of my pocket and be online in seconds.
I don’t think about how it works.
But somebody had to bury the fiber-optic cables.
Somebody had to build the cell towers and install the servers.
The phone companies had to upgrade their networks.
From 1995 through 2002, worldwide investment in telecommunications infrastructure totaled roughly $1.5 trillion.
Remember, this was 25 years ago. The entire world economy was much smaller than it is today.
And all that spending eventually gave us the internet infrastructure we barely think about anymore.
That’s the part I keep coming back to.
These technologies don’t magically appear. Somebody has to pay to build them.
And here’s the funny thing about artificial intelligence.
We think of AI as software.
But building it requires an incredible amount of physical stuff.
You need the chips and the data centers to put them in.
Those data centers need enormous amounts of electricity, which means more power plants, transformers, and copper.
Then you’ve got cooling systems, networking equipment, construction, land and everything else that goes into keeping the whole thing running.
None of this is cheap. And the bigger AI gets, the more of this stuff we’re going to need.
Barron’s recently looked back at some of the largest technology and infrastructure booms in American history and noticed something interesting.
During several of these giant buildouts, the amount eventually spent reached the equivalent of roughly 25% of annual U.S. GDP.
They called it the “Rule of 25.”
It’s not actually a rule. There’s no magic alarm that goes off at 25%.
But I think it’s useful perspective.
Today, the U.S. economy is more than $30 trillion. A quarter of that is around $8 trillion.
And even that might not be the right comparison.
The railroad numbers we’re talking about were American. But AI isn’t just an American buildout.
The chips are being made around the world. Data centers are going up all over the place. Countries everywhere are adding power and computing capacity.
And the world economy today is well over $100 trillion.
Now, I’m not saying we should take 25% of global GDP and decide that’s how much will be spent on AI.
That’s not how this works.
I’m just trying to put these giant numbers into perspective.
If we’re really building the infrastructure for a technology that could eventually touch almost every industry in the world, I have no idea what the final bill should be.
Is $1 trillion a lot?
Of course it is.
But is it a lot compared with what ultimately needs to be built?
That’s a different question.
Everybody Wants Your Money
I wrote recently about how Everybody Wants Your Money.
This is exactly what I mean.
If trillions of dollars are going to be spent building AI infrastructure, that money has to go somewhere.
Everyone immediately thinks about the chip companies.
But somebody has to provide all that electricity. Somebody has to build the data centers and cooling systems. Somebody has to supply the copper and networking equipment.
And somebody has to finance the whole thing.
That’s a lot of money flowing through a lot of different industries. And that’s where this gets interesting for us as investors.
Because I don’t need to know exactly how much money ultimately gets spent.
I want to know where the money is going. And then I want to see if it’s showing up in the stocks.
Here’s the part I think gets lost in all the bubble talk.
People can be completely right that too much money is being spent and still be way too early calling the end.
That happened with the railroads.
It happened with the internet.
People spent too much money. Companies went bankrupt. Investors got crushed.
But we still needed the railroads. And we definitely still needed the internet.
The fact that people eventually went crazy didn’t mean the technology wasn’t real.
It meant they eventually spent more money than the economics could support.
Those are two different things.
And that’s why I’m not looking at today’s giant AI spending numbers and automatically assuming we’re near the end.
Maybe we are.
But history gives me no reason to assume that just because the numbers sound ridiculous.
In fact, it makes me wonder the opposite.
What if the data centers we’re building today are the equivalent of the first railroad tracks?
What if the amount of electricity we’re talking about today looks small five years from now?
What if $1 trillion of spending sounds quaint someday?
I don’t know. Nobody does.
That’s why we’re going to keep doing what we always do.
We’re going to follow the money. We’re going to watch the stocks.
And we’re going to keep looking for opportunities in the companies getting paid to build all of this.
Because what looks like an enormous AI boom today could very well turn out to be something else.
The beginning…