AI Can Change the World and Still Disappoint Investors

“AI will transform the economy” and “some AI investments will lose money” are compatible claims. The missing link is the distribution of economic value: who creates it, who pays for it and who ultimately keeps it.

Follow the benefit, not just the spending#

A business buys computing services because it expects an operational benefit. That benefit might become lower prices for its customers, higher employee output or higher profit. The company supplying infrastructure receives only the portion captured in its contract.

The NBER discussion of generative-AI economics offers context on adoption and complementary investment. It does not imply that all resulting gains belong to listed AI suppliers.

Competition can transfer the payoff#

Consider a hypothetical industry where a new tool cuts production cost by 15%. If competitors all adopt it and reduce prices, customers may enjoy the improvement while industry profit margins remain similar.

Alternatively, a company with a differentiated service might retain more of the savings. The analysis must explain the mechanism that protects that profit, rather than assuming technology spending creates durable pricing power.

Capital costs determine the shareholder result#

Suppose a hypothetical project costs $100 million and generates $6 million annually after operating costs but before financing and taxes. Its usefulness does not tell us whether the return compensates investors for its risks, remaining investment needs and opportunity cost.

The BIS 2026 annual report examines investment competition and uncertain payoffs. Our broader conclusion is that an industry's economic importance does not eliminate the possibility of excessive investment.

Put the railroad analogy in its proper place#

The analogy is valuable because it encourages separate questions about infrastructure, customers and owners. It becomes misleading when surviving infrastructure is treated as proof that original shareholders did well.

For an AI stock, identify the expected profit pool, the company's defensible share of it, the funding needed and the valuation already assigned. A persuasive answer needs all four. Society's gain, on its own, is not an equity valuation model.

More in this analysis series#

Start with AI stocks versus railroad mania. Continue with AI Stocks vs Railroad Mania: What the Bubble Comparison Gets Right.

Analysis published October 1, 2026. Historical comparisons and hypothetical examples are not forecasts. Educational information, not investment advice.