Google is taking another major step toward challenging Nvidia’s dominance in the AI industry.

 

On Tuesday, Google announced that TorchTPU, its open-source version of PyTorch, is now available to developers. The move follows Amazon’s similar announcement and signals growing competition to reduce Nvidia’s control over the AI hardware market.

 

PyTorch is one of the most widely used open-source frameworks for building, training, and running AI models. Originally developed by Meta in 2016 and later transferred to the Linux Foundation, it became closely associated with Nvidia GPUs because many of its core features were designed to work with Nvidia’s hardware.

 

Google’s TorchTPU aims to change that by allowing developers to run PyTorch applications on Google’s Tensor Processing Units (TPUs). Developers can migrate existing applications to Google Cloud with minimal changes, potentially avoiding the need to rewrite their code.

 

Amazon is pursuing a similar strategy with TorchNeuron, which enables PyTorch applications to run on its custom-built Trainium AI chips and AWS infrastructure. Amazon is also contributing code to the main PyTorch project to make it less dependent on Nvidia-specific hardware.

 

Google is expanding its efforts by collaborating with companies and projects including Meta, Hugging Face, vLLM, and SGLang to improve compatibility with popular AI development tools. It has also introduced a TPU agent suite to help developers migrate applications, troubleshoot errors, and adopt Google's infrastructure.

 

However, easier migration does not guarantee that developers will abandon Nvidia GPUs. Nvidia's established ecosystem, software support, and developer familiarity remain significant advantages.

 

The bigger picture: Google and Amazon are competing not just to build AI chips, but also to make their hardware easier for developers to adopt. By reducing dependence on Nvidia's ecosystem, both companies are positioning themselves to capture a larger share of the growing AI infrastructure market.