Amazon Web Services released TorchNeuron on Monday, a free, open-source tool that lets developers run PyTorch code on Trainium chips with a single configuration change and no code rewrite, the company’s Annapurna Labs unit announced, according to IDNFinancials.
The software targets the CUDA ecosystem that has made Nvidia GPUs the default choice for AI workloads. About 85% of AI applications are built on PyTorch, according to AWS, and Nvidia has spent more than a decade optimizing CUDA for that framework. TorchNeuron is adapted from PyTorch itself, so developers who know the framework already know how to use it, TechCrunch reported.
More than 30 early customers — spanning robotics, world models and voice AI — are using the new software, said Colin Brace, vice president of software for Annapurna Labs.
“This has been about meeting customers where they’re at,” Brace told Reuters.
Upstream push
AWS is also contributing changes upstream to the main PyTorch project to remove code that ties the framework to specific chips, the company said in its announcement blog post. AWS aims to make PyTorch more hardware-agnostic for the entire developer community, not just Trainium users.
Cost and cloud integration
Brace said Trainium handles both training and inference on the same platform, which improves economics across a full workload. Deepgram, an AI voice technology provider, moved to Trainium using native PyTorch support and is processing 70% more audio per dollar at production-grade accuracy, Brace told TechCrunch.
Announcing TorchNeuron: @PyTorch code now runs natively on AWS Trainium chips with a single configuration change. No rewrite needed. Open sourcing at #PyTorchCon on Oct 20: https://t.co/6mz9jVhHTa pic.twitter.com/J3gxbtkkF7
“You’re building on AWS. You’re not betting on a single-purpose compute provider and scrambling to build everything else around it,” Brace said. “On AWS, the infrastructure to go from prototype to production is already there, allowing a founder to move fast and scale when they need to.”
AWS positions Trainium as a competitor to specialized cloud providers such as CoreWeave, Nebius and Firmus, which have deep relationships with Nvidia-based infrastructure, TechCrunch reported.

Competitive context
Google is also working to lure developers from Nvidia with better PyTorch tools for its custom AI chips, Moneycontrol reported. AMD and Apple already allow PyTorch users to port apps to their hardware with simple configuration changes, TechCrunch noted.
Encouraging developers to embrace Trainium without sacrificing the ease of use that has made PyTorch popular will be the tougher task for Amazon, TechCrunch noted. Eliminating hardware-specific requirements may provide greater flexibility but may also create extra configuration problems.
FAQ
What is TorchNeuron?
TorchNeuron is a free, open-source developer tool from AWS that lets PyTorch applications run on Amazon Trainium chips with a single configuration change and no code rewrite.
When was TorchNeuron announced?
AWS announced TorchNeuron on Monday, October 19, 2026, at a developer conference in Silicon Valley.
How does TorchNeuron challenge Nvidia?
Nvidia’s CUDA software platform has long been tightly integrated with PyTorch, making Nvidia GPUs the default hardware for most AI developers. TorchNeuron offers a comparable ease-of-use path to Amazon’s custom chips, and AWS is also pushing upstream PyTorch changes to reduce the framework’s automatic dependence on Nvidia hardware.
What performance claims has AWS made?
AWS cites Deepgram, which reports 70% higher audio-per-dollar performance on Trainium at production-grade accuracy using native PyTorch support.
Annapurna Labs, Amazon’s semiconductor division behind the Trainium chips used in this release, was established in Israel in 2011 by Hrvoje Bilic and Nafea Bshara and acquired by Amazon in January 2015. It now operates as a wholly owned subsidiary and is one of the largest fabless semiconductor companies in the world and a top-five customer of TSMC.
The unit has delivered seven generations of Nitro chips that power networking and storage for every AWS server since 2015, five generations of Graviton processors used by 98% of the top 1,000 AWS customers, and three generations of Trainium machine-learning chips, with Trainium2 powering Anthropic’s Claude models.
Amazon just launched TorchNeuron, which lets PyTorch code run natively on its Trainium AI chips. Here's what you need to know. TorchNeuron registers AWS Trainium as a native PyTorch device. It's built on the same extension points the community created to make PyTorch… https://t.co/BTlkJ1jts5 pic.twitter.com/VsORJSbyaN
Is TorchNeuron the only way to run PyTorch on non-Nvidia hardware?
No. PyTorch users can already port applications to AMD GPUs and Apple chips with simple configuration changes, though performance characteristics vary.
More on Annapurna Labs
We're open sourcing TorchNeuron. Your existing @PyTorch code runs on Trainium with one config change, and we're contributing upstream to make PyTorch more hardware-agnostic for everyone. https://t.co/Kpq6s2kz81 pic.twitter.com/TZS5NeCcJl