Why Everyone Is Wrong About the AI Bubble
Dec 23, 2025 · 21:43
Maxinomics asks if the AI boom repeats the Dotcom bubble, arguing it rests on three claims — AI keeps getting better, we need more data centers, everyone is using it — tested against the Dotcom era's 'critical lie': $500 billion of fiber, 90% unused as 'Dark Fiber.' It explains how models train on billions of word-guess flashcards, why one video frame packs 1 million times more information than a word, and why Nvidia chips run around the clock — no 'Dark GPU' equivalent — while data centers need power equal to 100 nuclear plants. It covers ChatGPT's retention rebound past Google's 95% benchmark, DeepSeek's $6 million model that cut Nvidia 17% overnight, and Jevons' Paradox — whale oil to kerosene — driving a 50x AI-use surge, concluding AI is a forest expanding, not a bubble.
