Double, Double Toil and Trouble

AI is already transformative, but the capital spending boom behind it is also creating the conditions for a speculative bubble and eventual industry consolidation.

• Previous transformative technologies followed this pattern: U.S. railway companies peaked at 1,564 in 1907, including 130 Class I companies, versus about 600 today and only six Class I operators; revenue rose from roughly $2.6 billion to $75 billion, while employment fell from about 1.5 million to 150,000.
• The U.S. automobile industry contracted from roughly 200 manufacturers in the 1910s to 44 by 1927, with Ford, Chrysler and General Motors emerging as leading players; today about 20 U.S. OEMs remain, while the former Big Three represent only 15–18% of a global market of almost 100 million vehicles.
• The dot-com bubble illustrates the investor risk: after the bubble burst in spring 2000, the Nasdaq fell more than 80% and took 15 years to recover its March 2000 high, even though internet infrastructure proved transformational.
• AI is already producing measurable benefits, including lower costs, millions of hours saved and productivity gains, but the industry could spend roughly $5 trillion on data centres and chips over five years while model builders remain at best marginally profitable.
• Infrastructure demand may be ahead of realized customer demand: current power requirements of roughly 40GW are being matched against plans for about 150GW of capacity, enough to power between 100 million and 124 million homes; AI racks can use up to 60 kW versus roughly 10 kW for traditional server racks.

Takeaway: Investors should distinguish the lasting productivity gains of AI from the profitability of the companies and infrastructure financing the boom; long construction times, debt-funded spending and excess capacity could turn today’s shortages into a glut, leaving a smaller group of winners after consolidation.