← All stories
Money

Mark Cuban predicts overbuilt AI data centres could end up ‘pickleball courts’

All-In Podcast · Mark Cuban: “A lot of data centers will be turned into pickleball courts.” · July 27, 2026
Mark Cuban predicts overbuilt AI data centres could end up ‘pickleball courts’
All-In Podcast
All-In Podcast
Mark Cuban: “A lot of data centers will be turned into pickleball courts.”
"There's going to be a lot of data centers that are going to be turned into pickleball courts. What's happening now is the market leaders, Google, etc. Meta, they're borrowing hundreds of billions of dollars."
Billionaire investor Mark Cuban argues that big tech’s AI build-out risks overcapacity, warning that data centres being built today could be stranded if efficiency breakthroughs sharply reduce power and compute needs. He says market leaders including Google and Meta are borrowing heavily while also directing cash flow into capex, and describes this as “planning for perfection” amid what he calls an existing private credit problem. Cuban compares the current moment to the fibre-optic boom that produced “dark fibre” sold for “pennies on the dollar” once capacity outpaced demand.

About this episode

In this short clip from the All-In Podcast, Mark Cuban forecasts that parts of the current AI data-centre construction boom could become redundant if rapid technology improvements reduce the compute and power required to run advanced AI systems. He describes a financing dynamic in which market leaders such as Google and Meta are, in his telling, borrowing “hundreds of billions of dollars” while simultaneously allocating most of their cash flow to capital expenditure, layering debt on top of already heavy spending.

Cuban frames this as “planning for perfection” and warns that a private credit problem already exists, implying additional borrowing could heighten fragility if expected utilisation or returns fail to materialise. He argues that while demand for AI services may rise, it could be offset by efficiency breakthroughs that compress costs and infrastructure needs.

To illustrate the risk, he draws an analogy to the fibre-optic build-out of the late 1990s and early 2000s, when successive performance improvements and overbuilding led to surplus capacity and “dark fibre” being sold for “pennies on the dollar”. His bottom line: similar price-performance gains in AI could leave today’s newly built data centres underused or economically stranded.

Key takeaways

More stories More from All-In Podcast