Google has joined Amazon in attacking Nvidia’s stronghold with AI app developers

Google released TorchTPU as an open-source project, letting PyTorch apps run on its Tensor Processing Units, and on Google Cloud, through a configuration change rather than a rewrite. A day earlier Amazon released TorchNeuron for its Trainium chips and said it is submitting code to the main PyTorch project to decouple features from Nvidia hardware.

PyTorch is the toolkit most AI apps are built on, by some accounts 85% of them, and it grew up around Nvidia GPUs. When two of the largest clouds attack that default in the same week, the cost of leaving Nvidia starts to fall.

We're for this. Competition at the software layer is the healthiest kind of pressure on a market where one chipmaker has set the terms. Nvidia's lock-in has always been as much about code as silicon: features in PyTorch were built to work with its GPUs because that's what researchers had in 2016. Making a move to TPUs a configuration change rather than a rewrite goes straight at the switching cost.

Google's edge here, per TechCrunch, is in the ecosystem work. It says it worked directly with Meta, Hugging Face, Inferact, the maker of vLLM, and RadixArk, the maker of SGLang, so TorchTPU behaves the way developers expect with the tools they already use. It also shipped a TPU agent suite to help port apps and fix errors along the way, and updated LiteRT for on-device PyTorch apps. That's the unglamorous compatibility work that actually moves developers.

Amazon's move may matter more long term. Submitting code upstream to decouple PyTorch from Nvidia benefits every hardware maker, not just AWS. A vendor-specific fork helps one cloud; a more neutral core helps the whole field. I'm watching whether that code lands, and how the Linux Foundation-governed project handles it.

The caution is the one TechCrunch itself flags. Easy portability doesn't mean developers leave Nvidia en masse. Performance, tooling depth and habit are sticky, and a port that works isn't the same as a port that's fast and cheap. There's also a trade worth naming: escaping one chip vendor's gravity by landing in one cloud's proprietary hardware is a different dependency, not freedom. The ideal endpoint is PyTorch that runs well anywhere, and that's the version worth rooting for.

PyTorch was developed by Meta, then Facebook, in 2016 and handed to the Linux Foundation in 2022. Google announced TorchTPU's open-source availability on Tuesday, a day after Amazon's TorchNeuron release.