Google initiated a project internally known as “TorchTPU” to improve its custom AI chips' compatibility with the widely used PyTorch software framework. Reuters reported that the tech giant is collaborating with Meta, the original creator of PyTorch, to optimize performance and reduce the engineering barrier for developers switching from Nvidia's GPUs. The initiative aims to erode Nvidia's software advantage by aligning Google's hardware, which previously favored its own Jax framework, with the industry-standard tools used by most AI researchers.
