Google and Amazon Take Steps Against Nvidia's Dominance in Artificial Intelligence

By opening the PyTorch-based TorchTPU project to open-source access, Google, alongside Amazon, took a significant step toward breaking Nvidia's dominance in the artificial intelligence hardware market. This allows developers to migrate their existing applications to alternative chips without changing code.

Google announced on Tuesday that the long-awaited PyTorch version of TorchTPU is now available as an open-source project. With this move, the company joined a joint effort with Amazon against the status quo in the artificial intelligence industry, where most applications run by default on Nvidia chips or GPUs. PyTorch, one of the popular open-source toolkits used by artificial intelligence developers and startups to build, train, and run their models, is preferred by 85 percent of artificial intelligence applications according to some data.

Developed by Meta in 2016, PyTorch was transferred to the Linux Foundation in 2022. Google's TorchTPU release enables artificial intelligence applications to run on Google chips called Tensor Processing Units (TPU) and allows existing applications to be migrated to Google Cloud with a configuration change. In a similar move, Amazon had announced TorchNeuron for its own Trainium chip and AWS, and stated that it would contribute code to the main PyTorch project to increase hardware independence.