The AI Boom's Next Problem shifts focus from compute hype to underlying unit economics and infrastructure profitability constraints.

Economists increasingly recognize artificial intelligence as a significant macroeconomic story, yet raise persistent questions about whether today's business models justify current valuations. Nordea, the Nordic banking group, has outlined the core tension: the investments are enormous and the stakes are high, but the bearish case may be more compelling than prevailing optimism suggests.

The fundamental challenge lies in inference costs—the ongoing expense of operating AI models after training completes. This is where artificial intelligence diverges sharply from classical software. While deploying software to an additional user carries near-zero marginal cost, inference requires continuous computational resources. Though cheaper chips have improved economics, user demands are moving in the opposite direction. Longer prompts, more complex reasoning, and larger outputs are becoming standard. Nordea argues that these pressures keep inference costs stubbornly high even as underlying technology improves.

A second pressure comes from competition and market structure. No single company has established a durable competitive advantage. Users can switch between models with minimal friction, and open-weight competitors continue narrowing the performance gap with closed systems. Frontier models—those at the technological edge—can be displaced within months, making them best understood not as lasting assets but as infrastructure with unusually short useful lives. This creates a difficult environment for companies making massive capital expenditures on data centers and compute capacity.

Nordea does not argue that AI will fail to generate enormous economic value. The critical question is not whether AI succeeds, but rather who captures the returns. If model quality converges, prices decline, and inference becomes commoditized, artificial intelligence could transform the global economy while concentrating returns among a far smaller set of winners than current valuations suggest.