Google and Amazon are stepping up efforts to challenge Nvidia’s position in the AI market by making their custom-designed chips easier for developers to use. Both companies have introduced tools that help developers move workloads from Nvidia GPUs to their chips using PyTorch, a popular open-source AI framework.
Google has open-sourced TorchTPU, a native PyTorch backend for its Tensor Processing Units (TPUs), developed in collaboration with Meta. Amazon has launched TorchNeuron, which lets PyTorch applications run on its Trainium and Inferentia chips with minimal code changes.
The tools aim to reduce friction for developers and reliance on Nvidia’s proprietary CUDA software platform, a key factor in the company’s market dominance.