Video
$930 Billion on AI Data Centers: Bigger Than Apollo. Is a Crash Coming?
Pete Cooper compares the roughly $930 billion spent on AI data centres over six years with projects such as the Manhattan Project, Apollo and the US railroads. He explains why he is uneasy about who owns that infrastructure and the three signs he thinks point to a coming correction.
Video summary
Pete Cooper says capital spending on AI data centres has reached about $930 billion over the last six years. He sets that against earlier megaprojects in today's money: about $36 billion for the Manhattan Project, $252 billion over 14 years for Apollo and about $500 billion over 71 years to build the US railroads. Unlike those, he notes, this infrastructure belongs mostly to a small number of hyperscalers, which he says are spending to capture the market and make AI cheap enough for more people and more uses.
He names two concerns. The first is that a handful of people on company boards control this infrastructure and, unlike democratic governments, cannot be voted out. The second is the rate of spending, which he says looks like a mathematical singularity, a curve that keeps getting steeper. He then tempers that with his experience of boom and bust cycles, most of all the dot-com crash around 2000, and calls a similar crash a very real possibility.
Pete sees three warning signs. Communities are pushing back on data centres over power, water and land. What he calls retail AI, the chat tools people use every day, is heavily subsidised to win users and data, and that money cannot flow forever. And while he sees many examples of AI helping companies, he says too much compute goes on things that are not useful, such as memes. He expects a correction over the next few years and hopes it turns out to be a moderate slowdown rather than a recession. Stressing that he is not a financial adviser, he suggests business owners and investors watch where the investment goes and consider redeploying their money and energy if a slowdown shows.
Video transcript
Read the spoken content without loading the YouTube player.
Did you know that over the last 6 years data centers capital expenditure has been 930 billion dollars? Now that's a number. I'm going to tell you what that means, why it's important, why it's important for you, and what you need to be aware of. First of all, I'm going to read you a couple of numbers and I encourage you to have a look at this graph and I'll flash it up here as well. The Manhattan Project. Now this was a government project started in World War II to develop the atomic bomb. The total amount in today's money was 36 billion dollars. America was first with the atomic bomb, used the atomic bomb a few times, and that helped to end World War II. I mean, a pretty big deal, right?
Let's talk about another project. In the Cold War, the Apollo project here, 252 billion over 14 years. All right? Again, another massive government project to put a man on the moon, US government project because of the Cold War to show space leadership and show the dominance of the US over the Soviets. Just compare that to that 930 billion dollars. Now have a look at that chart again. You'll see it's 6 years, it's multiple times as much. Let's have a look at another dimension, another couple of really big projects that are on this chart. The United States railroads, 500 billion, about half, but spent over 71 years to develop the railway infrastructure in the United States.
So again, huge investment over many, many years, combination of government and and private ventures there. And then in the Interstate Highway System, 62 billion in 37 years. So again, another government-funded infrastructure project. Now data centers are infrastructure. But who owns that infrastructure? Well, it's predominantly a very small number of hyperscales, big companies. And there's American, Chinese, and there's even some other countries that are dabbling in this as well. But predominantly, all of that capital expenditure is going into the creation of these data centers. Why are they pouring so much into it? What does it really mean?
Well, they all recognize that they're trying to capture the market. They're trying to create the infrastructure, the driving engine that makes AI cheap enough, more access to more people, more use cases, more features, and have us all using AI more and more. To make that work, you need a lot of data center infrastructure because it's predominantly processing power and data that runs the whole system, and that has to run on these data centers. A concern of mine is a few things. One is the consolidation in the ownership of that infrastructure.
It's not governments anymore, it's companies controlled by a handful of people, a company is built around a structure where we have a board of directors and a CEO, a very small number of people are making these decisions, and they don't go away. They can be incumbent in there for the rest of their lives. Unlike most democratic governments who can get voted out and new ones coming with new fresh ideas, and the old incumbents can go out. So, that's a concern that I have. The second big concern I have is the rate of spending in this area. If you're not familiar with a concept called the singularity, it's it's surprising how looking at that curve looks like a singularity.
So, in mathematics, a singularity is something that goes off to infinity. The graph just keeps going up and up, steeper and steeper. And surprising how that graph looks just like that singularity, that movement towards singularity. Is that a sign that we're moving towards singularity? Now, singularity has another context, and that is that we're heading towards an uncontrolled situation, a situation we don't understand where AI has effectively taken over, and we are no longer in control of our own destiny. The human race is no longer in control of its destiny.
Surprising how that CapEx growth looks very similar to, you know, a mathematical singularity and the correlation to, you know, this unknown future that we're heading into. So, all of that said, what does it mean? I mean, are we going to head for a singularity? Well, I actually would like to temper this a little bit and say I've been witness to many cycles in my career, cycles of boom and bust. The most memorable one would be the dot-com crash around the 2000s, where there was an enormous amount of money going into dot-com companies and then it crashed. It crashed dramatically and there were people sitting on hundreds of millions of dollars that were literally wiped out overnight.
Do I think that's going to happen? I think that's a very real possibility. We're going to see the hype that's pushing this up. It can't go up forever. Perhaps it can't go up forever like that and there's a correction. Where are the signs of corrections coming? >> >> Well, the the problems that rolling out data centers are starting to face. Pushback from the community, power problems, water problems, land problems. People don't want data centers in their back door, right? In in their backyard. They want them somewhere else. Everybody wants the data somewhere else.
We want the access to the features and the tools and the capability, but we don't want to have the infrastructure of the data centers. So, that's going to be a correcting factor. Also, we know, and if you look at my previous podcast, we know that AI is largely being subsidized. What I call retail AI where you're using a chat, it's heavily subsidized. So, you're not paying for all the AI usage and all of the compute that you're getting. You're getting it at a significant discount. Why are they discounting it? Because they want you to use it. They want to get the data. They want to get their flywheels working. They want to capture their market.
To do that, they've got to keep the prices low and in some cases free so that you're now using it and you're adopting it and it's becoming real for you. You're becoming reliant on that company, that that tool. So, when does the money run out? Do we keep pumping it in infinitely? There is never an infinite amount of money. So, somehow there has to be a correction coming. And I think the other thing is, well, where is the practical use of AI? Where does it help a company make money or lose money? How does it help people in everyday life?
I think there's a lot of promise, and I'm certainly seeing many, many examples that our companies using and developing now that does help companies, definitely help companies. But it doesn't help when people are using the compute to create useless memes like a cat with a donkey's head or whatever it is. That is not useful, and it doesn't help. And so we're seeing a lot of use cases that are not worthy of of anyone's time or energy. And that is third major factor. I believe we're going to see a correction. We're going to see a correction over the next few years, and the fallout from that could be pretty dramatic.
We could see recession or or even worse, and then it's going to be How How do we get climb back out of that? Hopefully I'm wrong. I hopefully things are are more of a moderate slowdown, but I do see a sense of recklessness that is going to have consequences. So what I suggest people do is is keep an eye on, you know, where this investment is is going, where it's starting to see problems. And if you're in business or you're investing, and this is not advice, I'm not a financial advisor, it's just my thought.
I would keep an eye on things, and if things start to show signs of a slowdown, I would seriously consider redeploying my investments and redeploying my energies and not following the hype. Um we have a podcast called Hype or Help. Please check that out, and we talk about hype and help. And most most of our podcast guests say that it's very helpful, but they're business people who are using AI to help them work better, more efficiently, uh do more, or do things more cheaply. But that's a long way from this massive spend that we're seeing on the hyperscalers and have capex. I hope that helps, and thanks for watching.
