Google Open-Sources TorchTPU, Challenging NVIDIA's AI Ecosystem Dominance

October 21 — According to the official Google Cloud blog, Google announced the open-sourcing of TorchTPU at PyTorch Conference 2026. The project was jointly developed by Google and Meta's PyTorch team, and the code has been published on GitHub. TPU thus becomes a native PyTorch hardware option alongside GPU and CPU, and existing code can be migrated with only minor modifications. Frameworks such as vLLM, SGLang, TorchTitan, VeRL, and Ray have already been integrated. In Hugging Face Transformers v5, passingdevice_map="tpu" allows models to be deployed to TPU. Tests show that TorchTitan training Qwen 3 8B achieves a maximum MFU (Model FLOPs Utilization) of 52%, and can scale to 8,192 Ironwood TPUs. Google also released llm-d 1.0 together with Red Hat and IBM, and LiteRT has also joined the PyTorch ecosystem.