# Google explains how Pathways coordinates large-model training with a single client

> Original title: What is Pathways? Single-controller training for large models

- Source: Google Developers Blog
- Published: 2026-11-09T00:00:00.000Z
- Bad Influence: https://badinfluence.news/items/klaed7l8x1bgu0gyv1cdb0tct
- Original: https://developers.googleblog.com/what-is-pathways-single-controller-training-for-large-models/

## Summary

Google's developer blog explains Pathways, the distributed system it built to train its own large models including Gemini, and announces its General Availability on Google Cloud Platform. The post describes how one client declares a training job as a graph, while the distributed runtime places the work, moves tensors between devices, and keeps accelerators busy. It contrasts this with SPMD execution, where lockstep collectives make pipeline stages and RL loops harder to overlap and a single machine failure restarts the run.

## Why it matters

The post contrasts SPMD training with Pathways' single-client model, giving a concrete framework for reasoning about coordination in multi-stage large-model training.