On Monday, Amazon Web Services released TorchNeuron, a tool that lets AI startups move PyTorch-based AI applications to AWS’s Trainium chip with a one-line configuration change and no code rewrite. The release targets the software advantage behind Nvidia’s dominance: CUDA, the infrastructure that lets AI frameworks such as PyTorch run on Nvidia hardware. Amazon says roughly 85% of AI apps are written with PyTorch, and TorchNeuron is adapted from PyTorch, so developers who already know PyTorch are expected to know how to use it.

Porting PyTorch apps to alternative hardware is not new; PyTorch users can also move apps to AMD GPUs or Apple chips with a simple configuration change. AWS argues the stronger draw is the rest of its cloud. Building on Trainium means developers are on AWS rather than on a neocloud tied to Nvidia and its hardware, such as Coreweave, Nebius, and Firmus. Colin Brace, VP of Technology Software at AWS Annapurna Labs, said: “You’re building on AWS. You’re not betting on a single-purpose compute provider and scrambling to build everything else around it.”

AWS also claims a cost benefit. Because Trainium handles both training and inference and is tightly integrated with the cloud infrastructure startups need, Brace says Trainium’s value is price-performance plus one platform for both workloads. He cited Deepgram, which moved to Trainium using native PyTorch support and is processing 70% more audio per dollar at production-grade accuracy. AWS does not claim the cheapest per-token price.

Beyond TorchNeuron, AWS says it is contributing changes upstream to PyTorch itself to remove code that ties the framework to specific chips, with the goal of making PyTorch more hardware-agnostic for the entire developer community. PyTorch’s purpose includes hiding the difficulties of getting AI code to run on hardware, so it remains to be seen whether this gives developers more choice and affordable options or creates more configuration headaches. Brace frames the bigger issue as removing the lock-in that kept developers from having a real choice at all.