Google has released TorchTPU, an open-source version of PyTorch, as a direct challenge to Nvidia’s dominance in AI development. The announcement follows Amazon’s release of TorchNeuron a day earlier, and both companies are trying to make their homegrown AI chips viable alternatives to Nvidia GPUs without forcing developers to rewrite their applications.

PyTorch is the most widely used toolkit for building, training, and running AI models, with some estimates putting its adoption at 85% of AI apps. It was created by Meta in 2016 and later handed to the Linux Foundation in 2022. Because Nvidia GPUs had long been the standard hardware for AI research, much of PyTorch‘s code was written specifically for Nvidia, giving the chipmaker a structural advantage that Google and Amazon are now trying to dismantle.

TorchTPU lets existing PyTorch apps run on Google’s Tensor Processing Units (TPUs) and on Google Cloud with a simple configuration change and no rewriting. Amazon’s TorchNeuron similarly ports PyTorch apps to Trainium and AWS. Amazon is also submitting code to the main PyTorch project to decouple features from Nvidia chips, making PyTorch more open to any hardware by default.

Google says it worked directly with Meta, Hugging Face, Inferact (maker of vLLM), RadixArk (maker of SGLang), and others to ensure TorchTPU works well with these tools. It also launched a TPU agent suite to help developers migrate and debug, and updated LiteRT to work better with PyTorch AI apps. None of this means developers will automatically abandon Nvidia, but Amazon and Google are clearly building the software bridges needed to make alternative chips and clouds a realistic choice.