History's capital cycle casts the AI boom as a railway moment

Every boom tells itself this time is different. In a June 29, 2026, column, Larry Swedroe drags that claim back through history. His guide is Edward Chancellor, the financial historian whose capital-cycle framework was built on the ruins of railways and dot-coms. Chancellor's warning is not that AI will fail as a technology. It is that the investors who fund it are the ones usually destroyed.

The cycle has a simple shape. High returns attract capital and competitors. Too much infrastructure gets built. Returns collapse, capital flees, and balance sheets get wrecked. Eventually the survivors inherit an industry with little competition, returns recover, and the whole thing starts again. Swedroe notes the same machinery has operated across mining, shipping, banking, real estate, and technology.

The cycle persists, Swedroe writes, not because markets are irrational but because of human tendencies. Overconfidence leads managers to believe their projections are sound. Competition neglect means that even as a manager invests to meet demand, they rarely ask what other managers are simultaneously building.

The sums at stake are historically unusual. The numbers involved in AI, Swedroe writes, dwarf anything the world has seen before. His column relays an interview that Kai Wu — host of 'The Intangible Economy' podcast and manager of Sparkline Capital's Intangible Value ETFs ITAN and DTAN — conducted with Chancellor.

Chancellor's framework leaves a sharp split between the technology and its financiers. The railways of the 1860s, Swedroe recalls, were supposed to bind a continent together and make fortunes for all; instead, the investors who paid for them absorbed the losses. Infrastructure remains and new users prosper, while the funders who arrive late in the cycle carry the wreckage.

Swedroe's column does not make a timing call or a valuation call. Its framework puts the risk on late capital. For an advisor, that frames a question the column does not answer: how much of the AI thesis is already priced into the holdings clients own. The answer, not the prediction, is where an advisor's control begins.