AI build-out needs $3.5 trillion a year by 2032 to break even

The setup is familiar from past technology cycles. In the late 1990s, telecom companies spent heavily on fiber optic networks on the assumption that internet traffic would grow fast enough to earn a huge return. Usage did explode, but the telecom companies overbuilt, new investment collapsed, and their stock prices took a big hit. The quality of equipment improved and costs fell, which helped internet-based businesses become financially feasible much more quickly — the users of the infrastructure, not its builders, captured much of the value.

Nineteenth-century railroad expansion followed a similar path. Companies invested heavily in materials and new routes at considerable cost, then competition and over-construction hurt the return on that spending. But shipping rates dropped significantly and rail became the cheapest and fastest way to move goods, directly benefiting factories and manufacturers.

That pattern applies to AI. If competition and technological improvements make AI substantially cheaper and more accessible, end users that implement it — manufacturers cutting production costs, transportation companies improving logistics, industrial firms using predictive maintenance — could become more attractive. The tech sector could still prove to be a good investment, but a lot of good news is already priced in, and these booms tend to reverse at some point. For now, those involved in the AI build-out are still benefiting; in time, the companies that best take advantage of what has already been built could be the biggest winners.