Amazon Web Services released a new developer tool on Monday called TorchNeuron. The tool lets AI startups move their apps onto Trainium, Amazon’s own AI chip.
The launch targets the software advantage that has helped Nvidia become the default hardware choice for AI apps.
TorchNeuron works with PyTorch, a free and open source toolkit that developers use to build, train, and run AI models. Amazon says apps written with PyTorch can run on Trainium by changing just one line of configuration.
According to Amazon, no code rewrite is needed. Because TorchNeuron is adapted from PyTorch, developers who already know PyTorch can use it.
Why PyTorch Matters in the AI Chip Race
Amazon says about 85% of AI apps are written with PyTorch. Facebook first developed the toolkit, and it has been part of the Linux Foundation since 2022.
Nvidia has spent years making sure its CUDA software supports PyTorch apps smoothly. CUDA is the software layer that lets apps run on Nvidia GPUs and other Nvidia hardware.
As a result, Nvidia hardware is deeply built into the toolkit and is used essentially by default.
Amazon is not the first company to try this approach. PyTorch users can already move their apps to AMD GPUs or Apple chips with a simple configuration change.
Amazon Pitches Its Full Cloud to Startups
Amazon argues that moving to Trainium also gives startups access to its entire cloud. AWS says this is a better choice than neoclouds such as CoreWeave, Nebius, and Firmus, many of which are tied to Nvidia hardware.
“You’re building on AWS. You’re not betting on a single-purpose compute provider and scrambling to build everything else around it,” said Colin Brace, VP of Technology Software at AWS Annapurna Labs. Annapurna Labs is the unit that makes Amazon’s custom chips.
AWS says Trainium handles both training and inference. Training teaches an AI model, while inference puts the finished model to work, such as answering a user’s question.
The company says this setup will save startups money, even though AWS does not offer the lowest per-token price.
“Trainium’s value is price-performance plus the fact that it’s one platform for both training and inference,” Brace said.
He pointed to Deepgram, an AI voice app provider that competes with ElevenLabs. Brace said Deepgram moved to Trainium using native PyTorch support and now processes 70% more audio per dollar at production-grade accuracy.
Amazon is also looking beyond its own hardware. The company said it is contributing changes to the main PyTorch project to remove code that ties the framework to specific chips.
Amazon said the goal is to make PyTorch more hardware-agnostic for all developers, not only Trainium users. It is not yet clear whether the change will give developers cheaper options or create more setup work.
“The real story is removing the lock-in that kept developers from having a real choice at all,” Brace said.
