AI Isn't A Bubble, But AI Spending Might Be.

Have you ever asked yourself whether today's AI boom reflects real demand or whether investors are simply confusing infrastructure spending with sustainable earnings growth?

Jim Chanos points to a striking historical parallel. Between mid-1998 and mid-2000, S&P 500 earnings rose roughly 30%, an acceleration from the growth rates seen earlier in the decade, convincing investors that a new era had arrived. Then, despite a relatively mild recession, S&P earnings collapsed by 40% within just twelve months. The reason was surprisingly simple: companies realized they did not need 10,000 routers after all, they needed 2,000. Order books were pulled, revenues evaporated, costs remained and profitability imploded almost overnight.

What makes the comparison unsettling is that the earnings collapse between 2000 and 2001 was as severe as the decline experienced during the Global Financial Crisis, despite the underlying recession being dramatically less severe. The problem was that investors mistook a capital spending boom for genuine demand - telecom companies had built enormous amounts of capacity because everyone believed internet traffic was exploding, supported by a widely accepted narrative that traffic was doubling every three months. In reality, research later showed it was doubling roughly every year, which was still impressive but nowhere near enough to justify the infrastructure being built.

Chanos argues that today's AI boom contains remarkably similar mechanics. S&P 500 earnings estimates continue moving higher because hundreds of billions of dollars in AI-related capital expenditures are flowing directly into the revenues and profits of a relatively small group of companies. The market rewards that growth with higher valuations because growth always looks attractive in the rearview mirror. Now a large portion of that growth depends on continued spending rather than proven long-term demand.

The critical question therefore is not whether artificial intelligence succeeds. The internet succeeded. The question is whether companies are once again building capacity based on assumptions that prove too optimistic. If spending slows because customers discover they need fewer chips, fewer data centers or less infrastructure than initially expected, earnings can deteriorate with extraordinary speed. History suggests that the most dangerous bubbles emerge when a genuinely transformative technology becomes attached to unrealistic expectations, because investors stop asking what is driving the growth and start assuming the growth itself is proof that everything is working.