Sep 23, 2026 · 30 min
Businesses reject costly frontier AI as investors chase expansion
Frontier AI is getting rejected by businesses | Dean Baker
The episode examines whether AI infrastructure spending can produce sustainable returns when companies favor cheaper, private, and customizable alternatives.
- 1Enterprise buyers often prefer older or open models because they deliver adequate performance at lower cost and with greater privacy.
- 2Frontier AI companies and hyperscalers face pressure from falling prices, concentrated customers, and data-center investments whose returns remain uncertain.
- 3Institutional incentives can keep money flowing into an AI bubble even as flawed automated analysis creates immediate systemic risks.
Don't miss
Dean Baker uses an unchecked analytical error that nearly triggered a serious military incident to argue that current AI risks deserve attention now.
The brief
Dean Baker argues that frontier AI depends on investors and customers supporting extraordinary infrastructure costs, leaving the industry exposed if enthusiasm fades before revenue catches up.
The conversation turns to falling token prices from OpenAI and Anthropic, raising questions about whether providers and hyperscalers can earn adequate returns on expanding data-center capacity.
Businesses may favor open-weight or older models because they are cheaper, customizable, more reliable, and better suited to keeping sensitive information private.
Baker says most companies want practical automation rather than superintelligence, while a small group of sophisticated customers may be capable of switching providers or adopting open models.
The episode’s sharpest warning links AI’s financial risks to its safety risks: an unchecked analytical error nearly triggered a serious military incident, Baker says.
Investors and pension funds can keep financing the boom because missing gains feels safer than challenging inflated valuations, even when broad economic returns remain unproven.