Believe it or not, 72% of the tokens generated account for only 27% of the money. Open-weight AI is looking like the volume layer of the AI market. New OpenRouter data for last week: 🔓 Open-weight → 71.8% of tokens → 26.9% of spend 🔒 Closed-weight → 28.2% of tokens → 73.1% of spend Oddly fascinating split (near mirror image). At the aggregate level, that works out to closed-weight traffic generating roughly 6.9× more spend per token than open-weight traffic. 🔓 Cheap/open models → volume 🔒 Expensive frontier models → premium intelligence And this may be where Local AI fits (yeah, Theo 🙂). 🎯 Saying it again, a local model doesn't have to beat Opus 5.5 at everything, it just needs to handle enough of the everyday workload cheaply, privately and fast enough AND send the really hard jobs to the frontier models. ⚠️ OpenRouter usage is cloud API usage. 72% open-weight doesn't mean 72% local or self-hosted spend, just to be clear. Image from OpenRouter