Amazon Launches TorchNeuron, a PyTorch-Based AI Tool as an Alternative to Nvidia

Amazon announced a software tool called “TorchNeuron” to facilitate access to AWS Trainium chips used in developing and running artificial intelligence applications. The tool, which enables PyTorch-based AI models to run on Trainium processors without rewriting existing code, was developed for businesses utilizing AI infrastructure, including e-commerce and retail companies.

Amazon Provides Trainium Support for PyTorch Developers with TorchNeuron

TorchNeuron, introduced by Amazon Web Services (AWS), works directly with PyTorch, an open-source artificial intelligence development platform. According to the company, approximately 85% of AI researchers and developers use PyTorch. With TorchNeuron, developers can transfer their existing applications to Trainium processors by changing just one configuration setting. Comparative tests conducted by the company reportedly showed that models trained on different processors achieved the same learning outcomes and training speeds.

The tool, which was announced to be released as open source under the Apache 2.0 license, was scheduled to be made available to developers during the PyTorch Conference North America, held on October 20, 2026.

An Alternative to Nvidia's Software Ecosystem Is Being Developed

TorchNeuron is part of efforts to develop an alternative to the CUDA software infrastructure that supports Nvidia's position in the artificial intelligence market. Nvidia's CUDA technology makes it easier for PyTorch applications to run on the company's graphics processors. AWS, meanwhile, aims to incorporate software changes into the open-source project that will reduce PyTorch's dependence on specific hardware.

Colin Brace, Vice President of Technology Software at AWS Annapurna Labs, stated that AWS infrastructure provides startups with integrated services ranging from prototype development to commercial deployment. Brace noted that using Trainium for both training AI models and running trained models provides cost advantages.

Artificial Intelligence Infrastructure for E-Commerce and Retail Companies

TorchNeuron is compatible with cloud infrastructure that can be used to develop applications such as AI-powered customer service, product recommendation systems, demand forecasting, and automated content generation in the e-commerce and retail sectors. AWS states that using the same processor infrastructure for training and inference operations allows resources to be utilized across different workloads.

In an example shared by the company, speech technology developer Deepgram reportedly increased the amount of audio processed per dollar by 70% after transitioning to Trainium.

Amazon Expands Open-Source Integrations

TorchNeuron supports Trainium2 and Trainium3 processors. The software, which integrates with Hugging Face development tools, also works with the TorchTitan platform, used for distributed training of large AI models. Amazon also contributes to the vLLM project by sharing open-source software components that accelerate the execution of AI models. The company reports that it tests each new version of TorchNeuron on hundreds of thousands of AI models and aims to expand developers' access to different hardware options.