TorchTPU is Google’s open-source project for running PyTorch applications on its Tensor Processing Units without rewriting application code.
Announced on 20 October 2026, it follows Amazon’s TorchNeuron announcement for moving applications to Trainium.
With PyTorch used by about 85% of AI applications, these projects show how major chip platforms are working to support an established framework.
Google and Amazon tools
On October 20, 2026, Google announced that it was open-sourcing TorchTPU. The tool allows PyTorch applications to run on Google's TPU chips. The company collaborated with Meta, Hugging Face, Inferact, and RadixArk to develop it. It also introduced tpu-agent-suite, which helps port applications to its TPUs.
A day earlier, on October 19, 2026, Amazon announced TorchNeuron. The tool helps port applications to Trainium, Amazon's chips.
Amazon is also contributing code to the core PyTorch project, with the goal of making its functions independent of Nvidia chips.
READ ALSO
OpenAI adds chat locking with PIN or biometrics to ChatGPT
Why PyTorch matters
PyTorch is a tool for building artificial intelligence applications and is used by about 85% of these applications. It was originally created by Meta in 2016 and is currently maintained by the Linux Foundation.
Google and Amazon tools make it easy to use their own chips from PyTorch applications without requiring rewriting the code, giving developers more choice in which chips to use.
G
Hacks.gr on Google
Set it as a preferred source for cybersecurity updates.
Preferred source
