Google has joined Amazon in attacking Nvidia's stronghold with AI app developers

What happened

Google announced TorchTPU as an open-source project. It lets existing PyTorch apps run natively on Google TPUs with a config change, no code rewrite needed. Amazon made a similar move the day before with TorchNeuron for Trainium chips.

Context and impact

An estimated 85% of AI apps use PyTorch, whose features were historically optimized for Nvidia GPUs. TorchTPU and TorchNeuron reduce the switching cost to alternative hardware, eroding Nvidia's technical moat.

Details

• TorchTPU: Google worked with Meta, Hugging Face, vLLM (Inferact), and SGLang (RadixArk)
• Includes a TPU agent suite to help developers migrate apps and fix errors
• TorchNeuron (Amazon): contributions to PyTorch core for hardware-agnostic defaults
• Analysts caution: developers will not automatically abandon Nvidia — ecosystem advantage remains strong
• Amazon is also selling Trainium chips externally (~$50B market opportunity)