Paul Kedrosky argues that AI infrastructure spending has reached historic levels, surpassing previous booms in canals, railroads, and fiber optics, with tech now dominating bond markets and over half of data center funding coming from debt. He contends that these investments are riskier than they appear due to rapidly depreciating token prices, constant hardware upgrades, and a duration mismatch for investors. The discussion challenges the notion that cheaper tokens will simply increase demand, pointing to the deflationary pressure on the core product and the challenges of hardware repurposing. Kedrosky critiques the logic that massive investment is justified because AI is useful, calling it a fallacy, and rejects the 'call option on AGI' argument as a 'god of the gaps' premise. He predicts the bubble will burst due to an overdetermined set of factors, including macro events, deflating token prices, and debt-funded construction, potentially leading to a financial crisis. While the collapse in data center spending seems inevitable, Kedrosky suggests AI itself will persist as a utility, with companies like OpenAI and Anthropic earning utility-like returns.