Ben Horowitz: AI infrastructure is wrong and will be rebuilt, from models on down
_AustinCalvert_ · x · 2026-11-06
a16z co-founder Ben Horowitz on what to expect in AI over the next 8-12 months. He argues AI infrastructure is wrong today, echoing early cloud computing when everyone thought they could port existing software — instead storage, networking, OSes, virtualization and containers were all rebuilt.
- Workloads: a database query is predictable; an AI query could run for a week, and systems to manage such requests don't exist yet
- Chips: Nvidia GPUs were built for video games — power-hungry and low-yield for today's uses
- Models: training sucks down all the data on the internet, and he doubts that's where we end up
His conclusion: everything from the model on down will change. In the same interview he lays out five market theses: AI is a new computer bigger than the internet; there's local excess but the 1999 math doesn't apply; money can now buy leads that took a decade to build; customers can rebuild much of their own software; and AI's infrastructure will be rebuilt. He cites ElevenLabs (valuation doubled to $22B) as his example of a moat big labs can't copy.
More from AGI Musings
- Jeff Ladish on AI hacking, self-replication and automated AI R&D risk — JeffLadish · 2026-11-06
- Why AI writing feels annoying despite models winning literary competitions — cis_female · 2026-11-06
- Why AI writing annoys us: models can win literary prizes, the problem is style fatigue — cis_female · 2026-11-06
- METR estimates AI is already accelerating AI development by roughly 2x — harris_edouard · 2026-11-06
- Lightcone Commons Opens AI-Safety Grant Applications, Backed by Jaan Tallinn and Moskovitz — NathanpmYoung · 2026-11-06
- METR: Anthropic now improving models twice as fast, driven by AI-assisted R&D — Recoil42 · 2026-11-06