Google and Amazon Release PyTorch Versions to Challenge Nvidia's AI Chip Dominance
Google announces on Tuesday that TorchTPU, its version of the popular PyTorch toolkit, is now available as an open source project. The move lets AI developers run their apps on Google's Tensor Processing Units (TPUs) with a simple configuration change and no code rewriting, effectively opening a path from Nvidia GPUs to Google Cloud. PyTorch powers an estimated 85% of AI applications, making it a critical battleground in the AI hardware market.
Google is not acting alone. Amazon releases its own PyTorch variant, TorchNeuron, on Monday, enabling porting to its homegrown Trainium chip and AWS. Amazon goes further by submitting code to the main PyTorch project that decouples features from Nvidia chips, making the framework hardware-agnostic by default. PyTorch, originally built by Meta in 2016, has been tightly coupled with Nvidia GPUs since its inception.
Google appears to be one-upping its rival by partnering directly with Meta, Hugging Face, and makers of popular AI tools like vLLM and SGLang to ensure compatibility. It also launches a TPU agent suite to help developers migrate apps and fix errors, and updates its LiteRT tool for on-device AI. While developers are unlikely to abandon Nvidia en masse overnight, the cloud giants are clearly working to erode its long-standing advantage.