Google launches TorchTPU to rival Nvidia

Google has released TorchTPU, an open-source tool that lets AI apps built with PyTorch run on Google's own AI chips. It arrived one day after Amazon released a similar tool for its chips. If you build or buy AI software, it means a little more choice in whose chips run it.

What happened

On Tuesday, 20 October 2026, Google announced TorchTPU as an open-source project, TechCrunch reports. It lets AI apps run on Google's Tensor Processing Units, or TPUs. These are Google's own chips for AI work.

The report says developers can move an existing PyTorch app to TPUs with a simple configuration change. They do not have to rewrite the app. That also moves the app to Google Cloud.

Google also released a "TPU agent suite," per the same report. It is a tool that helps developers move apps to Google's chips and cloud and fix errors along the way. Google also updated LiteRT, its tool for running AI on phones, laptops and small devices, so it works better with PyTorch apps.

Key facts

• Launch date: Google announced TorchTPU on Tuesday, 20 October 2026, per TechCrunch.
• What it does: it runs PyTorch AI apps on Google TPUs with a configuration change and no rewrite.
• Open source: Google released it as an open-source project.
• Partners: Google says it worked with Meta, Hugging Face, the maker of vLLM, the maker of SGLang and others on features and libraries.
• Amazon's move: Amazon released TorchNeuron on Monday, 19 October 2026, so AI apps can be ported to its Trainium chip and to AWS without rewriting.
• Reach of PyTorch: by some accounts, about 85% of AI apps use it, says TechCrunch. That figure is not exact.

Background

PyTorch is a popular open-source toolkit. Developers and startups use it to build, train and run AI models, per TechCrunch. Meta, then called Facebook, created it in 2016. In 2022, Meta handed it to the Linux Foundation, a nonprofit that looks after many open-source projects.

When PyTorch began, Nvidia's graphics chips were already the standard for AI research. So many PyTorch features were built to work with them. TechCrunch says most AI apps are written to run on Nvidia chips by default.

That is the lock Amazon and Google are trying to loosen. TechCrunch says Amazon is also sending code to the main PyTorch project. The code separates some features from Nvidia's chips, so PyTorch would be more open to any hardware by default.

What it means for you

This mostly affects developers, startups and companies that pay for AI computing. If you build AI apps with PyTorch, you now have an easier way to try Google's chips. You would change a setting, not rewrite your code.

For everyone else, nothing changes today. The apps you use look the same. Our take: more choice in chips could help on cost, but no source says prices will fall.

Our take: this is worth watching, not acting on yet. Before moving a real app, test it on a small job and compare cost and speed against what you use now.

What to watch next: whether big AI teams actually switch. TechCrunch says none of this means developers will leave Nvidia in large numbers.

What happens next

No source gives dates for what comes next. Amazon says it is submitting code to PyTorch, so watch for that code to be accepted.

Our take: also watch for early reports from developers who test TorchTPU and TorchNeuron on real apps.

FAQ

What is TorchTPU?

It is Google's open-source tool that runs PyTorch AI apps on Google's TPU chips, per TechCrunch. Developers make the move with a configuration change.

Does this replace Nvidia chips?

No. TechCrunch says it does not mean developers will drop Nvidia in large numbers. It gives them another option.

What is TorchNeuron?

It is Amazon's similar tool, released on 19 October 2026. It ports AI apps to Amazon's Trainium chip and to AWS without rewriting, says the same report.

Sources

• TechCrunch