Field notes
Working notes on AI infrastructure, product engineering, and the small numbers that change the math.
2026-05-19
The honest cost of running an AI chatbot: self-hosted LLM vs cloud API at every scale
I almost paid for an idle GPU. Here's the math I should have done first — break-even points for self-hosted LLMs vs Claude, DeepSeek, and OpenAI APIs at every traffic scale.