Google and Amazon have released open-source tools that let developers run PyTorch workloads on their own AI chips. The aim is to make leaving Nvidia hardware easier, which targets the software ecosystem around CUDA.
Google and Amazon are going after the software barrier that protects Nvidia. Google announced on October 20 that TorchTPU is now open source, bringing native PyTorch support to its Tensor Processing Units.
Amazon Web Services has made a similar move. It launched the public beta of TorchNeuron, an open-source native PyTorch backend for its Trainium chips.
What the two tools do
Google says existing PyTorch workloads can run on TPUs with minimal code changes while keeping familiar APIs and development tools. TorchTPU also integrates with Hugging Face, vLLM, SGLang, Ray and TorchTitan.
AWS, for its part, says developers can move PyTorch code onto Trainium with a configuration change instead of rewriting their models around a separate framework. The software currently supports Amazon EC2 instances powered by Trainium2 and Trainium3 chips.
Why software matters in the chip contest
Nvidia has spent years building CUDA into machine-learning development, and migrating a workload to other hardware has traditionally required engineering work, compatibility testing and optimization. Those migration costs help create hardware lock-in.
Native PyTorch support aims to change that. If developers can move models between GPUs and TPUs without rebuilding large parts of their software stack, hardware choice could turn more on price, availability and performance.
Nvidia keeps its advantages
Nvidia still holds significant advantages. Its GPUs remain widely deployed across AI training and inference, and its CUDA ecosystem includes years of optimized software, documentation and developer knowledge. Google and Amazon are therefore unlikely to remove that advantage overnight.
However, both tools are open source. AWS says models trained on Trainium can be saved as standard checkpoints and moved to other supported hardware, which may reassure companies worried about swapping one closed ecosystem for another.
Source: Crypto Daily
Trading involves risk.
Risk Disclaimer
