Google and AWS open new paths for PyTorch beyond Nvidia
At PyTorch Conference North America 2026 this week, Google open-sourced TorchTPU, a native PyTorch stack for Google TPUs, now available The post Google and AWS open new paths for PyTorch beyond Nvidia appeared first on The New Stack .
At the PyTorch Conference North America 2026, both Google and AWS have opened new pathways for PyTorch beyond Nvidia GPUs by open-sourcing TorchTPU and TorchNeuron, respectively. TorchTPU is a native PyTorch stack for Google's TPUs, while TorchNeuron is a native PyTorch backend for AWS Trainium. These open-source offerings are designed to make it easier for developers to switch hardware, as Nvidia's CUDA software ecosystem has long dominated AI development.
However, achieving top-tier performance on these unfamiliar hardware platforms is not straightforward, as PyTorch notes that a simple code change can get models running on TPUs but may require more optimization for peak performance. Google aims to make its TPU software ecosystem more open and developer-friendly, while AWS seeks to make Trainium a more practical option for developers using PyTorch.
Both TorchTPU and TorchNeuron are still in progress; Google recommends further tuning and production-grade stacks for optimal performance, while AWS outlines the need for three modifications to adapt a model for Trainium2.
Written by urgent.news from The New Stack's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
