An AI Boom Can Have Benefits and Bad Investments

The Financial Times published two accounts of AI’s economic prospects on October 18: a survey of 94 economists pointing toward gains beyond the United States and China, and reporting that America’s $10 trillion AI infrastructure boom is starting to fray. One concerns expected benefits across economies. The other concerns the economics of building the capacity behind the technology.

The available summaries do not specify the survey’s forecast horizon or how the $10 trillion estimate divides among spending, commitments, and projected projects. Those limits matter. An expectation of eventual growth cannot establish when revenues will cover a particular project’s costs. Nor should the buildout figure be treated as money already spent. The global economy has not cosigned every server order.

Granted, the strongest optimistic case does not require every infrastructure investor to prosper. Useful AI could help firms produce more with fewer resources, and competition could pass some savings to customers. A builder that pays too much for capacity could lose money while businesses using that capacity gain. Investor disappointment and public benefit can occupy the same spreadsheet, in different columns.

Who keeps the savings?

Still, diffusion is not automatic. Firms need affordable access, workable tools, and the ability to change how employees perform tasks. If a few platforms control access to essential services, those providers could retain much of the value through their prices and terms. That is a conditional risk, not a finding established by the survey. Its significance depends on whether customers have credible alternatives and whether lower computing costs reach them.

Overbuilding presents a different possibility. Excess capacity could push access prices down and help adoption while hurting owners’ returns. Yet poorly located or unsuitable capacity might not become a useful bargain. The FT’s infrastructure summary flags the economics of the physical world without supplying enough detail to diagnose particular projects. Physical constraints can obstruct delivery without proving that the underlying technology lacks value.

Timing separates these outcomes further. In the near term, builders must pay for construction and operation. In the middle term, customers must turn access into useful work. Over the longer term, competition and commercial terms help determine who retains the gains. These are analytical stages, not the economists’ documented forecast periods. Success at a later stage does not retroactively make every earlier investment well priced.

The optimistic forecast would strengthen if adoption broadened across countries and smaller firms, effective access prices fell, and users recorded sustained productivity improvements. Persistent high charges alongside concentrated gains would support the platform-rent concern. Weak utilization and poor project returns could support an overbuilding diagnosis even while some users benefited. Watch those measures together; neither a survey response nor a construction bill can substitute for them.