Easy porting to PyTorch is not unique to Amazon. Developers can already move PyTorch apps to AMD GPUs or Apple's chips with a simple configuration change. AWS argues the difference is what surrounds the chip: Trainium handles both training and inference, and startups using it get access to Amazon's whole cloud rather than, in AWS's framing, betting on a single-purpose compute provider such as the Nvidia-tied neoclouds Coreweave, Nebius and Firmus.

AWS also claims cost benefits from that integration, even if it does not offer the cheapest per-token price. Colin Brace, VP of Technology Software at AWS's Annapurna Labs, cited voice AI firm Deepgram, which says it moved to Trainium using native PyTorch support and now processes 70% more audio per dollar at production-grade accuracy.

Amazon is going further than its own hardware, saying it will contribute changes upstream to PyTorch itself to remove code that ties the framework to specific chips, making it more hardware-agnostic for all developers. Whether that delivers more choice and cheaper options, or simply more configuration headaches, remains to be seen.