Google, Amazon Release PyTorch Forks to Break Nvidia's Grip — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Google released TorchTPU as an open-source project on Tuesday, letting AI apps run on its Tensor Processing Units with a simple configuration change and no code rewriting
- PyTorch is used by roughly 85% of AI apps and was originally developed by Meta in 2016 before being handed to the Linux Foundation in 2022
- Amazon launched its own version called TorchNeuron on Monday, porting AI apps to its Trainium chips and AWS with no rewriting required
- Amazon is also submitting code to the main PyTorch project to decouple features from Nvidia's hardware, making the framework open to any chip by default
- Google worked directly with Meta, Hugging Face, Inferact (maker of vLLM), and RadixArk (maker of SGLang) to ensure TorchTPU integrates with popular AI development tools
- Google additionally launched a TPU agent suite to help developers migrate apps and updated its LiteRT tool for phones and laptops to better support PyTorch workflows
Why it matters: PyTorch's deep ties to Nvidia's GPUs — baked in since 2016 — have given Nvidia a structural moat even as competitors offer cheaper AI silicon. With both Google and Amazon now offering drop-in PyTorch forks plus dedicated migration tools for their own clouds, the switching cost away from Nvidia drops sharply for the developers building 85% of AI apps.
Ask SkimNews
