On Monday, Amazon Web Services released TorchNeuron, a new developer tool designed to let AI startups easily move their applications onto Amazon's custom Trainium AI chip. By requiring just a single line of configuration change for apps written in PyTorch, Amazon is directly targeting the software advantage that has long made Nvidia the default hardware choice for artificial intelligence development. Nvidia has spent years ensuring its CUDA software seamlessly supports PyTorch, embedding its GPUs deeply into the ecosystem.
TorchNeuron adapts PyTorch so that developers can port their applications to Trainium without rewriting code, a strategy similar to how users port apps to AMD or Apple chips. Beyond ease of use, Amazon argues that Trainium gives startups access to its comprehensive cloud infrastructure, saving them money by handling both model training and inference on a single platform. Companies like Deepgram have already adopted the tool to improve audio processing efficiency.
In addition to supporting its own hardware, Amazon is contributing code upstream to the main PyTorch project to remove framework code tied specifically to Nvidia chips. The broader goal is to make PyTorch hardware-agnostic for the entire developer community rather than just Trainium users. If Amazon succeeds in uncoupling PyTorch from Nvidia, it could reshape the competitive landscape for AI hardware and cloud services.