# AI Capex Super-Cycle All large frontier lab companies acting individually rational yet collectively irrational. Each chasing 100% market share while together spending 400%+ in capex. [[Hyperscalers]] (Microsoft, Meta, Amazon, Alphabet, Tesla, Oracle) committing $400B+ pursuing compute supremacy. ## The Irrationality If Google offers Gemini for free (as they did with Google Docs nearly two decades ago) or an open-source model offers compelling good-enough performance, competitors may never grow revenues to compensate for their build-outs. Individual rationality: "If I don't invest, I lose market position" Collective irrationality: "We're all building way more than demand justifies" ## Historical Parallels Pattern rhymes with: - Fiber optic buildout (1990s-2000s) - Colocation facilities (2000s) - Early cloud infrastructure (2010s) Most gross margin accrues to toll-keepers—Nvidia and the hyperscalers themselves. Rest face misallocation on monumental scale, over-investment followed by surplus. ## Outcome Prediction AI capex super-cycle will be: - Gloriously useful for society - Brutally destructive for many investors funding it Users surely win. Investors may scatter as infrastructure nonetheless compounds. But fundamental capacity gets built, enabling next wave of innovation. This is classic boom-bust infrastructure cycle: society benefits from overbuilding, investors pay for it. ## Investment Implications Watch for: - Who owns distribution (Google, Microsoft) - Who owns picks and shovels (Nvidia, TSMC, SK Hynix) - Who's spending capex without clear monetization path - Open-source disruption risk to closed models Related: [[Future of Compute]], [[Data Center MoC]], [[Investing Learnings]], [[Alpha in Atoms - Reflection on the Physical Inversion]], [[When to sell]] --- Tags: #deeptech #systems