I Asked Grok Whether AI Capex Is a Bubble — Its Answer Was Sobering

BennyBop · reddit · 2026-08-27

A Reddit user shared the output of a 10-minute conversation with Grok about whether US AI capex could produce a bubble. Grok's synthesis: a meaningful bubble (or at least a severe overinvestment cycle) looks more likely, and the risk of a sharp correction with financial spillover is real.

Its arguments: cost disadvantage + share loss — US hyperscalers are betting hundreds of billions toward $1T on capturing premium-margin usage, while Chinese models compete on most practical tasks at a fraction of the cost, and visible routed token share fell from 70% to 30% in about a year; security/data-control preference — organizations handling sensitive data rationally prefer self-hosted open-weight models, many of the strongest being Chinese-origin; circular funding and leverage — hyperscaler capex, NVIDIA-style vendor financing, rising debt and private-credit vehicles, with returns still largely prospective.

Verdict: the tech is real, but extreme cost plus rapid share loss plus circular financing makes current commitments fragile; significant repricing in AI infrastructure is plausible, while a 2008-style systemic crisis is possible but not the base case. Note this is an LLM's analysis, not original research.

Original post →

I Asked Grok Whether AI Capex Is a Bubble — Its Answer Was Sobering

More from AGI Musings