Billionaire Mark Cuban Compares the AI Buildout to the Dot-Com Era Fiber Glut. Here's What That Could Mean for AI Stocks.
Key Points
An infrastructure construction boom in the late 1990s brought vast amounts of fiber optic capacity online.
After the dot-com boom turned to a bust, 90% of that new infrastructure sat around unused for years.
As AI models become more efficient, they may need far less hardware than companies are racing to build.
Mark Cuban recently warned that the current artificial intelligence (AI) build-out -- the race to build as many data centers as possible as quickly as possible -- has echoes of 1999.
The billionaire thinks the build-out rhymes with the race to lay internet-enabling fiber optic networks in the late 1990s -- a somewhat underdiscussed aspect of the dot-com boom and eventual bust.
Missed Nvidia in 2009? This Rare Signal Is Flashing Again. In 2009, a "Double Down" signal flashed for a little-known chipmaker called Nvidia. For the first time in years, that same "Total Conviction" signal is flashing for a company 1/100th the size of Nvidia. Continue »
So, given that AI companies now make up something like 40% of the value of the S&P 500 (SNPINDEX: ^GSPC), should investors be worried?
Fiber boom, fiber bust
A quick history lesson: In the late 1990s, telecom companies spent a staggering $500 billion -- much of it borrowed -- to bury fiber optic cable across the country and build out the network to support it. This was the critical physical infrastructure that allowed the U.S. to transition online.
In the heat of the moment, companies believed that internet traffic would double every 90 days. But predictions outpaced reality. Though there were a couple of years when traffic did double every three to four months, after that, growth moderated. Over a sustained period, real traffic grew at roughly a quarter of that pace.
When the tech bubble burst in 2000, the industry discovered it had built about 10 times more capacity than anyone needed. About 90% of that new fiber sat "dark" and carried no traffic at all for years. A number of high-profile bankruptcies followed.
Image source: Getty Images.
Why Mark Cuban thinks AI could follow the same path
There was a simple mismatch between how much fiber was built and how much was actually needed. A huge part of that gap came from a too-rosy view of how much demand there really would be, but Cuban doesn't think that will be the issue in the AI infrastructure boom. In his mind, the issue will involve technical innovation.
During the fiber boom, engineers learned to push far more data through each strand of existing cable than was possible when those cables were first laid. This was another major factor in the overbuild. We just got a lot better at using what was already there. Efficiency and technical innovation actually worked against the fiber companies.
Cuban says the same kind of thing could happen today in artificial intelligence. As AI models become drastically more efficient, they could eventually need less and less hardware to support the same workloads.
That could lead to a lot of dark data centers turning, as he put it recently on the All-In podcast, "into pickleball courts."
The bottom line
There are definitely some major differences between 2000 and 2026 that could mean we avoid a crash like the one that ended the dot-com bull era. The most important: Until recently, the AI infrastructure build-out was mostly funded with the massive cash flows of companies like Amazon, Microsoft, and Meta Platforms.
Still, I think this warning is worth taking seriously. It's an interesting angle that I don't think a lot of investors consider. AI doesn't have to flop for there to still be a massive overbuild. Demand could keep growing rapidly, but innovations in AI itself could make much of the infrastructure being developed now unnecessary.
Johnny Rice has no position in any of the stocks mentioned. The Motley Fool has no position in any of the stocks mentioned. The Motley Fool has a disclosure policy.