Google open-sources TorchTPU to run PyTorch apps on its own AI chips, a day after Amazon's similar move

Google on Tuesday released TorchTPU, its long-awaited version of PyTorch for its Tensor Processing Units, as an open-source project. PyTorch is the open-source toolkit that developers use to build, train and run AI models, and by some accounts 85 percent of AI apps use it. Meta created PyTorch in 2016, when Nvidia's 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 in 2022. TorchTPU lets PyTorch apps run on Google's chips, and therefore on Google Cloud, with a configuration change and no rewriting. Amazon made a similar announcement on Monday with TorchNeuron, which moves PyTorch apps to its Trainium chips and AWS without code changes. Amazon also said it is submitting code to loosen Nvidia's hold on PyTorch itself. TechCrunch frames both releases as attempts to weaken Nvidia's position as the default hardware for AI applications. For developers, the releases could make it easier to switch between cloud providers and chip vendors. For Nvidia, they target one of the main reasons customers have stayed with its GPUs.