Between August 6 and 17 we measured both halves of the capacity-utilisation ratio — registered supply AND capacity actually in use — on the handful of networks that publish both. The result is the kind of finding node counts are structurally unable to show:
- io.net: registered devices +23.9% (2,196 → 2,722). Compute-hours
- Akash: fleet shrank 5–7% while demand surged — GPU utilisation
A dashboard ranking these two by node growth inverts reality.
Full write-up with method, daily paths, and every caveat we could find in our own numbers (single-window disclaimer, intraday-swing handling, a frozen counter we excluded, a fill-rate we refuse to publish and why): the divergence report.
Every figure links to the network's own public endpoint. Check us in one click — that is the product. API access.