Google TPUv3 POD Generic is a large GPU cluster operated by Google, in the private sector. It came online around May 2019. At that moment it ranked number 18 among known clusters.
It is built on 1,024 Google TPU v3. That amounts to about 63.64 H100e of AI compute. The cluster draws roughly 1.02 MW. Its hardware is estimated at $10M.
Epoch AI rates the certainty of this record as confirmed. Refers to the general commercially available pod, not to a specific cluster using the pod