Amazon unveils TorchNeuron to bring PyTorch to Trainium

Amazon announced TorchNeuron on October 19, 2026, a software tool designed to let developers run PyTorch code on its Trainium artificial-intelligence chips with a configuration change rather than a code rewrite. The company says the tool could make it easier for developers to use its chips, which compete with Nvidia’s processors. Amazon engineers said TorchNeuron would enter public beta on October 20 and initially support Trainium2 and Trainium3. The software is intended to work with PyTorch, a widely used framework for building and training AI models. Amazon describes TorchNeuron as open source; its engineers said it would be released under the Apache 2.0 license. The tool addresses a significant barrier to switching chip platforms: software compatibility. Nvidia’s CUDA software supports AI applications running on its GPUs, and developers may face added work when adapting code for different hardware. Amazon says TorchNeuron connects PyTorch workloads to Trainium with a single device-setting change. The company also plans to contribute changes to PyTorch that it says would reduce framework code tied to particular chips. The release remains in beta, and the initial version has...