Driving past American commercial real estate reveals a pattern of overbuilding. Structures once built for specific purposes often find new uses when initial demand fades. This cycle of investment exceeds capacity is now playing out across multiple sectors.
Currently, hyperscalers like Alphabet (GOOGL), Microsoft (MSFT), Meta (META), and Amazon (AMZN) are channeling nearly $700 billion toward AI infrastructure, with data centers forming the bulk of these expenditures, according to CNBC.
The AI Infrastructure Crisis: A Case for Prudence
Cuban, who sold Broadcast.com to Yahoo for $5.7 billion in 1999—a deal that later stranded buyers—argues that the current AI data center spending mirrors past misallocations. Unlike the dot-com crash, however, this scenario involves physical infrastructure with long depreciation timelines.
His core concern is not AI’s viability but the efficiency of capital deployment. Even if AI adoption grows, breakthroughs in model efficiency or hardware could render vast portions of the constructed capacity obsolete.
The risk extends beyond equity markets. Data center construction drives utility infrastructure investments, including substations and power grids. These physical assets may not adapt quickly if demand does not materialize as projected.
Cuban suggests smaller IPOs for AI companies—targeting $50-100 million valuations—to broaden investor exposure and reduce concentrated risk. He emphasizes that current mega-deals concentrate downside risk among private equity and venture firms.
Three metrics underscore the scale: $700B in 2026 AI spending plans, $176B in potential depreciation overestimation through asset lifecycle stretching, and AI companies now comprising 34% of the S&P 500 by weight. Retail investors indirectly participate through index funds but face concentrated risk.
