Google TPU Pods
Google's unit of training: thousands of TPUs behind optical circuit switches.
You're looking at a system record: an assembled machine — a server or a whole rack — with what's inside it and where it's deployed.
Google scales TPUs in pods — 4,096 chips (v4), 8,960 (v5p), 9,216 (Ironwood) — wired by inter-chip interconnect in a 3D torus and reconfigured through optical circuit switches that route around failures without touching a cable. Gemini trains across multiple pods over the Jupiter network; the pod, not the server, is Google's hardware product.
Specifications & key facts
Each figure is labeled by basis: vendor spec, vendor claim, reported, estimate, or official. "Not publicly disclosed" means exactly that — we never invent values.
Overview
How this connects to the rest of the stack
This record's links at a glance — click a group to jump to it.
Designed by
Cooling
Contains
Sources
Google Cloud TPU architecture ↗
Record reviewed as of 2026-01-15 — see Methodology.