Tech & Science 22 Oct 2026
Google opens TorchTPU to loosen Nvidia's software grip
Google has open-sourced TorchTPU, software that lets developers run artificial intelligence work built with PyTorch on its Tensor Processing Units (TPUs) with few code changes, The Hindu reported. It was announced at PyTorch Conference 2026 on October 20. Two days earlier, Amazon released TorchNeuron for its Trainium chips. Both moves aim to loosen Nvidia's hold, since its CUDA software has made its chips the default choice.
Who it affects, and how
- AI developers they can try Google or Amazon chips without rewriting much code
What it means for your work
- Engineer / tech professional: Developers can move AI workloads to Google or Amazon chips with less rewriting.
- Founder / startup: Cheaper alternatives to Nvidia chips may lower AI computing costs.
What it sets off
Who decided, why they say they did, and what it changes for people. Tap a card for the reasoning.
Green edge: good for those affected; red: bad; grey: mixed. Reported and Expert view steps come from a named source; Our reasoning steps are not stated anywhere and follow through the mechanism shown, so we are less sure of them. Bars show how sure we are, and that falls with every step. How we reason
Who is in it
Where this comes from
A level-1 summary of these reports, written in plain words. We have not checked it claim by claim; stories we research in full are marked level 2 to 5.
- [1]Business News Today – Finance, Industry & Economy | The Hindu: After Amazon, Google targets Nvidia’s software lock-in with TorchTPU original ↗
- [2]Business News Today – Finance, Industry & Economy | The Hindu: Amazon targets Nvidia’s software edge with new AI tool original ↗