Transcription
The biggest tech companies in the world are pouring hundreds of billions of dollars into artificial intelligence, racing to build huge data centers as fast as they can. They see AI as the future, of course, and they don't want to fall behind. Nobody wants to miss out.
But despite all that money and urgency, nearly half of these data centers might never get built. On top of that, the AI bubble appears to be on the verge of popping. We talked about valuations before in my previous videos. Now, the reason is not a lack of demand or funding. Of course, it is much simpler and more surprising. They don't have enough electricity and they can't get the basic equipment that is needed to power these facilities. That's where the real problem begins.
Right now, the US is in the middle of a massive AI buildout. Companies are planning data centers that would add around 16 gigawatts of capacity by the end of 2026. But only a small portion of that is actually being built. Experts say between 30 and 50% of these projects are being delayed or cancelled. Now there are many issues related to this and I will go into further detail in an article that I'm writing and that I will publish on my Substack and Patreon.
But the first big issue is power. AI data centers use huge amounts of electricity, far more than older facilities. Some are so large that they use as much power as an entire city. One project in Texas, for example, will need about the same electricity as nearly a million homes. Now, imagine dozens of these being built at the same exact time. The power grid just isn't ready for that. It is not devised to handle that. And it's not just, um, you know, AI using electricity; electric cars, heating systems, and other new technologies are all competing for the same energy. So even if a company has the money to build a data center, it might not be able to turn it on.
Now there are many other issues, but the second problem is equipment. Data centers need things like transformers, switch gear, and batteries to work. These things are not exciting, but they are essential, and without them, nothing really runs. Transformers are especially important because they help move electricity from the grid into the data center. But right now, there aren't enough of them. There aren't enough transformers available. In the past, it took about 2 to 3 years to get a transformer. Now, it can actually take up to 5 years. That is just far too long when companies want to build data centers in just over a year.
To make things harder, the United States doesn't make enough of this equipment at home. It imports it. And this is where de-industrialization becomes a big problem. So, companies have to import it. And, uh, guess where they import it from? They often import it from China. That creates a very strange situation. It is a dependency, but Washington would admit it. Of course, the United States is trying to compete with China in AI, but it still needs, uh, parts from China. It still depends on China to build its own AI infrastructure. It sounds very ironic, doesn't it? So, because of trade tensions and supply chain problems, getting these parts isn't always easy. Projects get delayed, costs do go up, and timelines fall apart. Some companies are even reusing old transformers from shutdown power plants just to keep moving forward.
When you step back, the scale of investment is huge. Companies like Amazon, Microsoft, and Google are expected to spend more than $650 billion on AI infrastructure. But even that might not be enough to solve these problems that I just discussed.
This is where a bigger question comes in. Is the AI boom moving too fast? Behind the scenes, there are many issues and there are signs of strain. One major issue is something called circular financing. This means big tech companies invest billions into AI companies and then those AI companies spend the money right back on the same companies, um, buying chips, renting servers, and using their data centers. So effectively, it is a circle. So the money keeps flowing into a loop, and it helps boost growth and company values. Remember, valuations are skyrocketing, and they're basically based on just assumptions that are not grounded in reality. But it also effectively hides a problem. Many AI companies are not making profits. Their costs keep rising as their systems get more complex and expensive to run. And of course, they're driving investment valuations that are not based in reality. Because of that, they need constant funding just to stay alive, just to stay operational, just to stay a growing concern. And that raises an important question. What happens if the money slows down?
Some people inside the industry are already thinking about their risk, not because a collapse is certain, uh, although of course many of them do believe that the bubble is just about to pop, but because there are warning signs that they can no longer ignore. Spending is massive, returns are unclear or non-existent, and now even building the infrastructure is becoming challenging. It is becoming difficult or impossible if the trade war with China continues.
This doesn't mean AI has no future. Of course, the technology is real, and it will likely play a big role, a significant role in the years ahead. We know that it is being widely used in the military, and given where we're going, I'm assuming that Washington is going to invest significant funds into AI military infrastructure. Uh, but, uh, it does suggest that the current hype might be getting ahead of reality. At the end of the day, this isn't just about software or smart algorithms. And it's not just about building structures here domestically. It is about real-world limits such as electricity, access to resources, manufacturing, supply chains, and relations with those countries on whom you depend, such as China. And right now, those limits are starting to show, and some of them are being strained very, very significantly.
Nearly half of planned data centers are being delayed or cancelled, as I mentioned in the beginning of the video. And it's not because companies don't want them, but because they physically cannot build them fast enough. So the big question I would say is very simple. Is this just a temporary slowdown that will be fixed over time? Or is it the first real sign that the AI boom is starting to hit its limits? Because if the foundation can't really keep up, everything that is built on top of it is very uncertain. You would agree with that.
Thanks so much for watching. I will continue updating you on the AI bubble. And as I said during the video, there's going to be a post on my Substack and Patreon with more details where we explore the financial and the economic impact of the AI bubble. I appreciate you being here and I look forward to seeing you here again tomorrow. Take care.