Google Releases TorchTPU to Help Developers Move AI Apps Off Nvidia Chips
Developers can move existing PyTorch apps to Google Cloud TPUs with a configuration change instead of rewriting code. Announced October 20, 2026, TorchTPU was built with Meta, Hugging Face, Inferact and RadixArk as Google challenges Nvidia’s dominance.
Google announced on Tuesday, October 20, 2026, the release of TorchTPU, an open-source version of PyTorch that allows AI applications to run on Google’s Tensor Processing Units (TPUs) instead of Nvidia’s GPUs. The move positions Google alongside Amazon, which made a similar announcement the day before, in an effort to reduce Nvidia’s dominance in AI development infrastructure.
PyTorch is the open-source toolkit used to build, train, and run AI models, accounting for roughly 85% of AI applications by some estimates. Originally developed by Meta (then Facebook) in 2016, it was built at a time when Nvidia GPUs were already the standard hardware for AI research, meaning many of its features were designed specifically for Nvidia chips. Meta transferred the project to the Linux Foundation in 2022.
TorchTPU lets developers move existing PyTorch applications to Google’s TPUs and Google Cloud through a simple configuration change, with no rewriting required. Google says it collaborated with Meta, Hugging Face, Inferact (maker of vLLM), and RadixArk (maker of SGLang) to ensure compatibility with widely used AI development tools. Google also launched a dedicated TPU agent suite to help developers migrate their apps and resolve errors, and updated its LiteRT tool — which runs AI on phones, laptops, and embedded devices — to work better with PyTorch applications.
Amazon’s equivalent release, TorchNeuron, enables PyTorch apps to run on its homegrown Trainium chip and AWS without rewriting. Amazon additionally said it is submitting code to the main PyTorch project to decouple features from Nvidia’s chips, making the toolkit more hardware-agnostic by default.
Nvidia currently holds a strong default position in AI development, with most AI apps written to run on its hardware. These moves by Google and Amazon may lower the barrier for developers to switch to alternative chips and cloud platforms, though whether developers will do so at scale remains to be seen.