Full episode · 1 h 4 m

Google's AI Infrastructure Chief, Amin Vahdat, on the Physics & Economics of Frontier AI

Training Data · Sonya Huang & Pat Grady with Amin Vahdat

Share

The best minutes

1. 5:27Goodput, not FLOPS, is how Google holds itself accountable5 min
2. 18:12Google split the TPU into 8i and 8t based on market size4 min
3. 34:14Long-horizon agents push CPU, storage and network demand up4 min

Also worth watching

1. 26:10How Google and DeepMind change TPU designs before tape-out8 min
2. 43:32Why Google waits on utilities rather than building its own gigawatt4 min
3. 49:14Serving clusters cost more than training, and old TPUs stay fully used5 min

Skip the opening on data center basics and the Ineffable rack, the optical switching explainer, and the open standards segment.

Published Oct 6, 2026.