It lets AI apps run on Google's Tensor Processing Units (TPUs), the company's own AI chips.
Existing PyTorch apps can be moved to TPUs, and therefore to Google Cloud, with a simple configuration change and no rewriting.
A day earlier, Amazon released a similar product, TorchNeuron, which ports AI apps to its Trainium chip and to AWS.
Most AI apps are written by default to run on Nvidia's GPUs, and PyTorch, which developers and startups use to build, train, and run AI models, was shaped around them.
Meta created PyTorch in 2016, when Nvidia GPUs were already the standard hardware for AI research, so many of its features were built specifically for them.
Meta handed the project to the Linux Foundation, a nonprofit that manages major open-source projects, in 2022.
Making it easy to switch chips and clouds lowers a barrier that has kept developers tied to Nvidia, though that does not mean they will leave in large numbers.
Amazon is also submitting code to the main PyTorch project that separates various features from Nvidia's chips, so the toolkit works with any hardware by default. Google says it may go further in other ways: it worked directly with Meta, Hugging Face, Inferact (maker of vLLM), RadixArk (maker of SGLang), and others so TorchTPU works well with popular AI development tools. It also launched a TPU agent suite to help developers move apps to its chips and cloud and fix errors along the way, and updated LiteRT, its tool for running AI on phones, laptops, and embedded devices, to work better with PyTorch apps.