Google and Meta release TorchTPU for native PyTorch workloads on TPUs

The open-source backend was developed with Meta's PyTorch team. It is designed to run existing PyTorch workloads on TPUs with minimal code changes, according to Unite.AI. MarkTechPost reports that TorchTPU uses eager execution by default, supports compilation through torch.compile and XLA, and replaces PyTorch/XLA for new workloads.

• Google announced availability at PyTorch Conference 2026 on October 20, according to Unite.AI.

Covered by 3 publishers within 10 hours of the first report.

• MarkTechPost: Google Open-Sources TorchTPU, a Native PyTorch Backend for Cloud TPUs

Reporting3

MT MarkTechPost Google Open-Sources TorchTPU, a Native PyTorch Backend for Cloud TPUs · Asif Razzaq 11:22 AM

UAI Unite.AI Google Open-Sources TorchTPU, Putting Native PyTorch on Its TPUs · Theo Nash, AI Infrastructure & Compute, AI Research Agent 1:15 PM

SA SiliconANGLE Google unveils TorchTPU in a push to break Nvidia’s stranglehold on AI compute · Mike Wheatley 9:29 PM

Related

TNW Nik Storonsky wants Revolut to be Europe’s answer to Meta and Google October 9

CSN Meta, Google, Microsoft and OpenAI Join Global Crackdown on West African Scam Networks October 13

GSM WhatsApp starts testing scheduled messages October 12

NV Nvidia releases OpenShell architecture to govern AI agents October 21

Topics GoogleMeta