Google Open-Sources TorchTPU to Break Nvidia's Grip on AI Developers

Google open-sourced TorchTPU on Tuesday, its version of PyTorch that runs AI apps on Google's Tensor Processing Units — a direct attempt to break Nvidia's stranglehold on AI development.

PyTorch, by some accounts, powers 85% of AI apps. That figure matters because most of those apps are written by default to run on Nvidia's GPUs. Whoever controls the default path controls where developers deploy — and where they spend.

The stakes are straightforward. Google and Amazon want that spending redirected to their own chips and clouds.

Why PyTorch favors Nvidia in the first place?

Meta, then Facebook, originally developed PyTorch in 2016 — years before large language models reshaped the industry. Nvidia's GPUs were already the standard hardware for AI research at the time. Many of PyTorch's features were therefore built to work specifically with Nvidia chips.

Meta handed PyTorch to the Linux Foundation in 2022. That governance change created the opening Google and Amazon are now exploiting.

In short: an accident of timing gave Nvidia a decade of default status.

What does TorchTPU actually change?

TorchTPU lets AI apps run on Google's TPUs. It also lets developers move existing PyTorch apps to TPUs — and to Google Cloud — with a simple configuration change and no rewriting of their code.

That last point is the core of the pitch. Rewriting AI apps for new hardware has historically been the barrier that kept developers on Nvidia. Google claims that barrier is now a configuration file.

Amazon made a similar announcement Monday with TorchNeuron, which ports AI apps to its homegrown Trainium chip and to AWS with no rewriting. Amazon said it is also submitting code to the main PyTorch project that decouples various features from Nvidia's chips, making the framework more open to any hardware by default.

That submission, if accepted, could matter more than either company's standalone release.

How is Google going further than Amazon?

Google says it worked directly with Meta, Hugging Face, Inferact (maker of vLLM), and RadixArk (maker of SGLang) to ensure TorchTPU's features and code libraries work well with these popular AI development tools, just as the original PyTorch does.

The company also launched a dedicated tool, the TPU agent suite, to help developers move their apps to Google's chips and cloud and fix errors along the way. And it updated LiteRT, its tool that runs AI on phones, laptops and embedded devices, to work better with PyTorch AI apps.

The ecosystem partnerships are the notable move. Compatibility with tools like Hugging Face addresses the practical risk that a port works on paper but breaks in real workflows.

Will developers actually leave Nvidia?

Probably not immediately, and the source is candid about that. None of this means AI app developers will automatically ditch Nvidia's hardware en masse just because they can now easily move to alternative chips and clouds.

What has changed is the cost of leaving. A migration that once required rewriting applications now requires, in Google's telling, a configuration change.

Amazon and Google are clearly working to end Nvidia's advantage. Whether developers reward that effort depends on whether TPUs and Trainium deliver the performance and pricing to justify even a simplified switch.