adtestbench Higgsfield

AI news · Amazon · chips

Amazon opens its Trainium AI chips to PyTorch code

Amazon says TorchNeuron makes its chips a standard PyTorch target, so AI code can leave Nvidia hardware without a rewrite.

By Alexandre S. , 00:30 UTC

Amazon Web Services announced TorchNeuron on Monday 19 October 2026, a free add-on that lets AI code written for PyTorch run on Amazon’s own Trainium chips. PyTorch is the toolkit most developers use to build and run AI models. Amazon says about 85 per cent of AI apps are written with it, TechCrunch reports.

Retro-futurist illustration: a striped sunset over a grid horizon under a starry sky, with a rocket motif standing on the horizon.
Drawn by adtestbench from “Amazon opens its AWS Trainium AI chips to every PyTorch developer, no rewrite needed”,

The catch for chip rivals has always been the code. CUDA, the software layer that drives Nvidia’s chips, has been tuned for PyTorch for years, which is why Nvidia hardware is where most AI apps end up. In the basic example in the PyTorch announcement, the only change to a training script is to send the model to a device called “neuron”; compiled mode needs one more line. It covers Trainium2 and Trainium3, in the step-by-step mode researchers use and in the compiled mode used in production. The code goes public as a beta under the Apache 2.0 licence on 20 October, at the PyTorch Conference.

AMD and Apple chips already offer a similar simple configuration change, TechCrunch notes. AWS sells Trainium on price for the performance and on having the rest of its cloud in the same place.

For anyone paying per AI clip or image, more chips that run the same code give the providers behind those tools more room to compete on cost.