https://m.youtube.com/shorts/Rn0qsdB2ZTQ
Mark Cuban: “A lot of data centers will be turned into pickleball courts.”
116K views
From Pickleball Courts to Dark Fiber 2.0: Why the AI Data Center Boom Could Go Bust
The hottest real estate in America right now isn't beachfront property in Miami or a penthouse in Manhattan. It's a windowless concrete box in the middle of the desert, humming with servers, guzzling electricity, and costing billions of dollars to build. Data centers are the new gold rush, and every tech giant is racing to build more, bigger, and faster than the last. But what if this gold rush ends the same way every other infrastructure boom has ended? With empty buildings, cheap assets, and a lot of very expensive pickleball courts.
That striking image comes from a recent conversation about the current state of artificial intelligence investment. The warning was simple: there are going to be a lot of data centers that are going to be turned into pickleball courts. It sounds like a joke, but it is a serious prediction rooted in history, finance, and technology. And if you are paying attention to where the money is going, it is a prediction worth taking seriously.
The Scale of the Bet
To understand why so many people are worried, you have to understand the sheer scale of what is being built. Companies like Meta, Google, Microsoft, and Amazon are not just spending their profits on artificial intelligence. They are spending all of their cash flow on capital expenditures for AI infrastructure, and then borrowing hundreds of billions of dollars on top of that to spend even more.
This is not normal corporate spending. This is a bet-the-company level of investment predicated on one idea: that demand for AI computing will grow forever, at current prices, with current technology. It is what financiers call planning for perfection. Everything has to go right. AI adoption has to continue at a breakneck pace, customers have to be willing to pay premium prices for AI services indefinitely, and the technology itself has to remain as computationally and power hungry as it is today.
If any one of those assumptions breaks, the math falls apart very quickly. And there are already cracks forming in the financing that makes this entire boom possible.
The Private Credit Powder Keg
Much of this data center expansion is not being funded by traditional bank loans or even straightforward corporate bonds. It is being funded by private credit. Private credit is essentially lending that happens outside the regulated banking system, often with higher leverage, less transparency, and fewer safeguards.
There is already a growing concern among investors and economists that the private credit market is overheated. Companies have loaded up on debt in a period of higher interest rates, and many of those loans have not yet been tested by a downturn. Now, layer on top of that several hundred billion dollars in new borrowing to build data centers that may not generate a return for years, if ever.
When the biggest, most cash-rich companies in the world are borrowing heavily to fund speculative infrastructure, it creates a fragile system. Their cash flow looks strong today, but that cash flow is being entirely consumed by building. If revenue growth from AI slows even slightly, or if interest rates stay high, those debt payments become much more painful. It is a classic boom-time setup. Everyone is borrowing because everyone else is borrowing, and no one wants to be left behind. But when the cycle turns, the debt remains long after the hype fades.
The Jevons Paradox Trap
The most common counterargument to this bubble warning is simple: You don't understand AI demand. The need for compute is insatiable. Every new model requires more power, more chips, more data centers. Even if we make AI more efficient, we will just use more of it.
In economics, this is known as Jevons Paradox. When a resource becomes more efficient to use, we do not use less of it, we use more of it because it becomes cheaper and more accessible. Proponents of the data center boom argue this will absolutely happen with AI. Make it cheaper and more efficient, and people will just build bigger models and find new uses, consuming all the capacity we build and then some.
They are probably half right. Utilization will absolutely grow. The world will use more AI in five years than it does today, likely by an order of magnitude. But that does not mean we need all the data centers we are building today. And that is where the second half of the equation comes in: technological breakthroughs.
The Price-Performance Cliff
The entire financial model for today’s data centers is based on today’s technology. Today’s leading AI models are incredibly power hungry. They require massive clusters of GPUs running constantly, cooled by enormous amounts of water and electricity. A single large data center can consume as much power as a small city.
But what happens if the power requirement collapses? What happens if a breakthrough in model architecture, chip design, or algorithmic efficiency means you can do the same work with one tenth of the power and one tenth of the chips?
History tells us this is not just possible, it is inevitable. Technology does not stand still, especially in computing. The price-performance curve always moves in one direction: you get more performance for less cost and less power over time.
If that curve moves quickly for AI, the implications for data centers are profound. A facility designed to house 100 megawatts of power-hungry servers suddenly becomes wildly over-provisioned. You do not need that much power, that much cooling, or that much space anymore. The newest, most efficient models can run in a fraction of the footprint. The older, massive data centers become obsolete, stranded assets. They are too expensive to operate and too inefficient to compete. And you are left with a very large, very empty, very well-cooled building. Perfect for pickleball.
We Have Seen This Movie Before
If this sounds familiar, it is because we have seen this exact cycle before. In the late 1990s, during the dot-com boom, the mantra was fiber, fiber, fiber. The internet was going to need infinite bandwidth, so companies raised billions of dollars and laid fiber optic cable across the entire country, under oceans, and through every major city.
For a few years, it seemed like a brilliant bet. Then technology improved. The invention of dense wavelength division multiplexing, or DWDM, meant you could send vastly more data over a single strand of fiber. What once required ten strands of fiber could now be done on one. A 1 gigabit connection became 10 gigabits, then 100 gigabits, all on the same glass that was already in the ground.
Suddenly, there was not a fiber shortage. There was a massive fiber glut. Miles and miles of dark fiber, as it was called, sat unused in the ground. Companies that had spent billions to lay it went bankrupt. Investors who had funded the buildout lost everything. And a few years later, savvy buyers came in and bought that dark fiber for pennies on the dollar. The infrastructure was valuable eventually, but it was built too much, too soon, and with too much debt.
Every single person who lived through that era remembers the lesson. The demand was real. The internet did change everything. But the infrastructure bet was still wrong because it underestimated how quickly technology would improve.
Why AI Data Centers Are Next
The parallel to AI is almost perfect. Today, we are in the lay more fiber phase. We are building data centers as if power and compute efficiency will stay flat forever. We are assuming that the only way to get more AI is to build more giant, power-hungry boxes.
But innovation is already happening at a frantic pace. Chip companies are designing more efficient processors. Researchers are developing smaller, more powerful models that require less training. Techniques like quantization, mixture-of-experts, and inference optimization are slashing the cost to run AI. And entirely new computing paradigms are on the horizon.
It does not take a miracle to make today’s data center look inefficient. It just takes a few incremental breakthroughs compounded over three to five years. And when that happens, the economics flip. The company that can offer the same AI performance for half the price and a quarter of the power wins. The company stuck with a 10-year lease on a gigawatt data center filled with last-generation hardware loses.
No one is saying AI is not real or that it will not be transformative. It will be. Just like the internet was. But transformative technologies do not guarantee that every infrastructure investment made at the peak of the hype will pay off. In fact, history shows the opposite is true. The most transformative technologies create the biggest gluts, because everyone overbuilds at once.
The Pickleball Future
So what does the endgame look like? It is not that data centers disappear. We will always need them. But we will not need as many of them as we are currently building, and we will not need them in their current form. The most power-hungry, debt-fueled, hastily built facilities in the least strategic locations will be the first to become stranded.
They will sit empty, costing money to maintain, until someone finds a new use for a large, flat, climate-controlled box with a lot of power lines running to it. An indoor farm, a warehouse, or perhaps, a very expensive community pickleball court.
The market leaders will survive, just as the big telecom companies survived the fiber bust. But the investors, the private credit lenders, and the companies that over-leveraged to build at the top will pay a steep price. The warning is not that AI will fail. The warning is that success for AI does not mean success for every data center. And the more money we borrow to build for a perfect future that never arrives exactly as planned, the more pickleball courts we are going to have.
Responses « Index