Home
ProductsComfyUIMulti-cloud console
PricingBlogsDocs

Does PyTorch Run on Google TPUs?

Back
Team Aquanode

Team Aquanode

Sarthak Vaish

NOVEMBER 13, 2026

Yes. Google now points PyTorch users to TorchTPU, an open-source, PyTorch-native backend for TPUs. It replaces PyTorch/XLA, which is deprecated.

TorchTPU

Google's TorchTPU documentation describes the following:

  • Co-developed with the PyTorch team at Meta, and open source.
  • Eager execution by default, with one process per device.
  • torch.compile captures the graph with TorchDynamo and compiles it for TPU through OpenXLA and StableHLO.
  • Distributed training through torch.distributed with the tpu_dist backend, supporting DDP, FSDP2 and DTensor.
  • A tpu device in place of cuda: torch.device("tpu"), installed with pip install torch_tpu.

PyTorch/XLA is being retired

The PyTorch/XLA project page says "This project is deprecated and replaced by TorchTPU" and that the repository "will be archived on November 15, 2026." It asks teams to start new PyTorch TPU work with TorchTPU and move existing workloads over.

JAX-based serving stacks remain supported separately for JAX workflows. For the chip itself, see what a TPU is, and for the wider field, NVIDIA's main competitors in AI.

#pytorch#tpu#google#xla

Related reading

Does Anthropic Use Google TPUs?

Yes. Anthropic announced plans to use up to one million Google TPUs, and says it runs on three chip platforms: TPUs, Trainium and NVIDIA GPUs.

How Do You Access a Google TPU?

Google documents TPUs as a service you use through its own cloud products, and its TPU documentation describes no way to buy a chip.

Is a TPU Better Than a GPU for AI?

Only for some workloads. Google says TPUs suit matrix-heavy models, long training runs and large batches, while GPUs suit models with many custom operations.

Your own GPU cloud accounts

One console for the GPU clouds you already rent from. Your keys stay on your machine. $20 a month, flat.

Connect your accounts

One console for the GPU clouds you already use.

Connect your accounts for $20 a month, flat. Your provider bills the GPU.

Aquanode LogoAquanode

Run AI without spending any time on infrastructure.

© 2026 Aquanode. All rights reserved.

All trademarks, logos and brand names are the property of their respective owners.