Amazon's TorchNeuron lets AI apps switch off Nvidia chips

Amazon Web Services has released TorchNeuron, a free tool that helps AI apps move from Nvidia chips to Amazon's own Trainium chip. It came out on October 19, 2026. Nvidia has long been the default chip for AI apps, TechCrunch reported. Our take: a real choice could mean lower costs for AI startups.

What happened

AWS released TorchNeuron to help AI startups run their apps on Trainium, TechCrunch reported. The tool works with PyTorch, the free, open-source toolkit many developers use to build, train and run AI models. Per Amazon, an app written in PyTorch can run on Trainium by changing a single setting, with no code rewrite.

The tool is adapted from PyTorch, so developers who know PyTorch already know how to use it, TechCrunch reported. Amazon says its goal is to make PyTorch "more hardware-agnostic for the entire developer community, not just Trainium users," TechCrunch reported.

TorchNeuron is open source under the Apache 2.0 license and is hosted on GitHub, wpnews reported. It works on Trainium2 and Trainium3 chips, and a public beta went live the day after the announcement.

Key facts

• PyTorch share: About 85% of AI apps are written with PyTorch, by Amazon's own figure, per TechCrunch.
• Customer result: Voice AI company Deepgram moved to Trainium and now processes 70% more audio per dollar at production-grade accuracy, according to AWS, per TechCrunch.
• Earlier test: AWS first announced native PyTorch support through TorchNeuron as a private beta on December 22, 2025, per AWS.
• Debugging and speed: The tool supports step-by-step running for easy debugging, plus a compile mode for faster production use, wpnews reported.
• Not the first: PyTorch users can already move apps to AMD or Apple chips with a simple setting change, TechCrunch noted.

Background

Nvidia's CUDA software lets apps run on Nvidia graphics chips, and it has long supported PyTorch apps smoothly, TechCrunch reported. That has made Nvidia hardware the default choice for many AI builders. PyTorch was first built by Facebook and has been part of the Linux Foundation since 2022, TechCrunch reported.

Amazon contrasts itself with smaller "neoclouds" such as CoreWeave, Nebius and Firmus, many of which are tied to Nvidia, per TechCrunch. The same Trainium chip handles both training and running models. AWS says that saves startups money even if AWS is not the cheapest per token, per TechCrunch. An AWS technology vice president told TechCrunch, "The real story is removing the lock-in that kept developers from having a real choice at all."

What it means for you

Our take: most people will never touch this tool. But if AI startups can run on cheaper chips, the apps you use could get cheaper or add features faster. More competition for Nvidia is usually good news for buyers.

Our take: the proof will be how many startups actually switch.

What happens next

Our take: watch for more customer stories like Deepgram's, and for whether more developers move their PyTorch apps to Trainium.

FAQ

What is Amazon TorchNeuron?

It is a free, open-source AWS tool that lets PyTorch AI apps run on Amazon's Trainium chips by changing a single setting, per TechCrunch.

Does TorchNeuron replace Nvidia?

It gives developers another option besides Nvidia chips. PyTorch apps can already move to AMD or Apple chips with a simple change, per TechCrunch.

Which chips does TorchNeuron support?

Amazon's Trainium2 and Trainium3 chips, wpnews reported.

Sources

• TechCrunch
• wpnews.pro
• aws.amazon.com