Is the AI Buildout the Next Debt Crisis Trigger?

**Natalie Brunell** (0:00)
Sometime in the last day or two, you probably picked up your phone and asked AI a question. How to word an email? What to make for dinner with what's in the fridge? That's not the same as a Google search. A search looks up something that already exists. AI is building you an answer from scratch, one word at a time, and it didn't happen in your phone. Your question left the house, traveled through cyberspace, across the state or maybe even the country, and landed in a windowless building packed with humming computers.
Powerful chips inside processed your question and sent the answer back. That building is a data center, and we're spending an almost unimaginable amount of money building them. Morgan Stanley estimates $2.9 trillion in global data center spending between 2025 and 2028 That's more than the entire market cap of Bitcoin. Yes, it's one of the reasons money has rotated out of Bitcoin this year. But here's the part most people don't realize. Roughly half the money for this AI build out has to be borrowed or raised from investors. That raises a bigger question we'll come back to.
How long can this keep working if every new wave of AI spending requires even more money than the last? But first, let's see what all the money is actually buying. Here's what's inside these data centers. The thing everybody's been fighting over is a powerful chip called a GPU, a piece of silicon about the size of a stamp packed with billions of microscopic switches. And at the most basic level, its job is surprisingly simple. It does math, just really, really fast math, which sounds like it has nothing to do with writing you a sentence, right? But every word gets turned into a number first. That's what lets a machine do math on language. That's what companies like OpenAI with ChatGPT, Anthropic with Claude, Google with Gemini, and Musk's XAI with Grok are all doing. Building an AI model means feeding it basically everything humans have written, books, articles, websites, video transcripts, conversations, and having the model repeat the same exercise over and over. Guess the next word. Check if it was right, adjust, go again, billions and billions of times. Once the AI model is built, the work doesn't stop. Every time you ask it a question, those chips fire up to build you an answer. Now multiply that by hundreds of millions of people using AI now every day. That obviously takes an enormous amount of computing power, and all that computing has to live somewhere. So several of those chips go inside a powerful computer called a server, and then you stack thousands and thousands of those servers row after row until you have an entire building of them, and all those chips run super hot. So every datacenter needs enormous amounts of power and cooling, which is why a huge part of the money isn't even for the computers. It buys concrete, steel, power lines, transformers, cooling systems, and connections to the electrical grid. That's why the biggest datacenters are measured in gigawatts, a power scale you usually only hear about with major power plants, not campuses full of computers. So who's building all of this? Mostly it's a handful of enormous technology companies, Microsoft, Google, Amazon, Meta, and Oracle. They're called the hyperscalers. Some of them also make their own AI models. Google does, Meta does. Elon Musk's XAI built a massive site called Colossus in Memphis, Tennessee to power Grok. Other hyperscalers are basically landlords renting out the computing inside their datacenters to AI companies that need them. And the spending has just exploded. For years, these giant technology companies were so profitable, they could pay for all the expansion themselves. They could build new datacenters and still have cash left over. But now the buildout has gotten so expensive that that no longer is enough. And one example shows just how extreme the numbers have become. OpenAI's revenue is now running at roughly $25 billion a year. And it's looking at a massive new datacenter in Ohio. But NVIDIA has reportedly been in talks to provide roughly $250 billion in financing guarantees connected to that datacenter. That is 10 times OpenAI's current annual revenue. Now NVIDIA is the company making the AI chips these datacenters need. So why would NVIDIA help with the financing? Well, because if the project gets financed, the datacenter gets built. And if the datacenter gets built, it needs a huge number of NVIDIA chips.
So NVIDIA isn't just selling the technology going into the AI boom. In some cases, it's also helping its customers get access to the money needed to build it. And this month, the scale got even bigger. NVIDIA just recently announced partnerships with some of Wall Street's biggest firms, including BlackRock, Blackstone and Goldman Sachs, aimed at raising more than $500 billion to finance more AI infrastructure. So step back and look at what's happening. This is becoming much more than a technology story. It is a major story in finance. And that brings us back to the question from the beginning.