Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Borderline all or nothing bet. The distribution of AI capex investment and burnt/stranded assets is not like past bubbles.

Railroad boom. Most of spending moving heart to build out rail network that increase CONUS rail by +400%. US freight backbone. 50+ year infra. Upgrade signal etc and improve network capacity.

Dotcom Boom. Most of spending in moving earth to build out ~150m of fiber. Agnostic pipes for backbone of internet. Also 50+ year infra. Upgrade switches at nodes to increase network capacity.

AI boom. Most of investment going into chips with <5 year deprecation cycle. Datacenters <10 years, heavily specialized for AI use, i.e. power/cooling makes it inefficient/not economic for general compute.

Looking like ~20% goes to eletricfication capex (if it gets built at all). That's the stranded asset i.e. consolation prize, ~10% (generous) increase in US electrification, which isn't nothing, but also not substantial.

Also consider Tulip mania. Tulips die in a few weeks. All value gone.

TLDR potentially bulk of AI investment is closer to tulips in terms of deprecated / stranded assets past medium term i.e. <10 years, than rail or fiber.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: