How to Sound Like an Expert in Any AI Bubble Debate (www.derekthompson.org)

A new explainer — co-authored with Timothy Lee — distills the AI-bubble debate into a 12-item “playbook” of statistics, studies, and talking points people recycle in media and investor rooms. The piece walks readers through the strongest pro-bubble arguments (and counterpoints), aiming to equip anyone to follow or join conversations about whether AI investment is rational or a mania. The authors emphasize that some arguments are persuasive while others are overused or misunderstood, and they show where the evidence actually lands rather than just repeating headlines. Key technical details: five tech giants (Amazon, Meta, Microsoft, Alphabet, Oracle) reported about $106 billion in capex in a recent quarter — roughly 1.4% of U.S. GDP — much of it tied to AI data centers. High private valuations and massive seed rounds (e.g., Thinking Labs’ $2B seed and talks of a $50B valuation) raise red flags. METR’s experiment (16 programmers, 246 tasks) found developers thought AI sped them up ~24% but actually took ~19% longer, though other field studies (32 orgs using Cursor) report 26–39% productivity gains. The piece highlights worrying financial linkages — Nvidia’s “up to $100B” tie to OpenAI, an OpenAI–Oracle deal reported at ~$300B, and use of SPVs to move AI spending off balance sheets — creating potential circular financing and contagion risks if demand or valuations tumble.