Mark Cuban on the AI Bubble: Who Actually Gets Wiped Out?

The conversation starts by comparing the current AI boom with the dot-com era, which both participants say they lived through. One participant agrees there is "bubbly like behavior" but argues the shape is different. In the late 1990s, companies with "no revenue, no traffic, no nothing" went public at extreme valuations, their stocks rose 50% or 100%, and ordinary people talked about them in taxi cabs. The speaker says that kind of retail mania is absent today.

It's not the traditional dot-com bubble, right?

The claimed consequence is that any collapse would land on a narrower group. It would not hit "most people in the room" or most people across the U.S. Instead, the speaker says it "could just destroy a lot of VCs and a lot of funds and a lot of P.E." because those investors are "going all in."

The discussion connects this to competition among fund managers. They used to need to beat a benchmark such as the SPX. Now, according to the speaker, they also have to outperform the fund next door to keep money coming in. That pushes them into crowded positions in companies such as Anthropic and SpaceX, which works only until conditions turn. The other participant, who says they have done venture investing for just over 10 years, describes watching many investors who deployed capital "at the wrong time" go out of business after investing at the peak. They argue that entry price matters. As angels, they used to invest at $5 million or $10 million valuations. Later the asks became $40, $50 or $60 million for products that had not launched.

The second risk the speakers raise is leverage. Market leaders such as Google and Meta are described as borrowing hundreds of millions of dollars on top of spending their cash flow on capital expenditure. The speakers say this comes while there is "already a private credit problem." They also mention 50-year bonds, which they describe as planning for perfection.

The argument on data centers rests on efficiency. If AI's price-performance improves enough to cut power requirements, the speaker expects many data centers to become stranded. The other participant adds that some facilities simply cannot get power turned on.

there's going to be a lot of data centers that are going to be turned into pickleball courts.

The historical analogy offered is fiber. The build-out once seemed to be the whole story, but capacity jumped from one to ten to 100 gigabytes (as transcribed), the bandwidth problem disappeared, and dark fiber later sold "for pennies on the dollar." The speaker asks why AI and data centers would not see the same price-performance improvements. They also suggest that public hostility toward data centers might, in a sense, protect existing ones. They stress that long commitments of tens of billions of dollars over 10 to 20 years are hard to forecast: "nobody can predict that well."

The speakers do name the case that cuts against this. Demand for tokens from world models and video could absorb the capacity. One participant concedes: "If I'm going to be wrong on the data centers, it's going to be because of video." They link video to world models and robotics.

Building on the "pricing to perfection" idea, the discussion argues that a thesis such as OpenAI putting $100 billion to work must pay back in earnings and margin dollars, not just in revenue. "It has to be profitable," and the speakers say the outcome is unknowable.