Epoch AI

node

The research group that maintains the public databases this Journey depends on for anything quantitative about inputs: a catalogue of notable machine-learning models with reconstructed training-compute estimates, a benchmarking hub, and from 2025 a database of large AI data centres built from satellite imagery and permit filings. It matters to the argument of this Journey structurally rather than substantively. Almost every widely quoted figure about how fast AI compute is growing traces back to one small organisation reconstructing numbers that the developers themselves do not publish, and its own documentation is explicit that the estimates are reconstructions with stated uncertainty and partial coverage. A field whose central quantitative claim rests on third-party estimation of undisclosed inputs is in a different evidential position from one whose numbers are filed, audited or independently reproducible.

A node is not a place. Drawing it on a map would assert something about the world that no stored fact supports.