Amazon makes its biggest move against Nvidia yet with a new TorchNeuron development tool

Meta description: Amazon Web Services launched TorchNeuron to move PyTorch AI apps onto Trainium with one configuration change, challenging Nvidia's cloud lock-in.

Tags: Amazon Web Services, TorchNeuron, Trainium, PyTorch, Nvidia

Featured image: Abstract cloud and neural network chips connected by one configuration line, no people, no logos, no text.

Monday Launch: A One-Line Bridge From PyTorch To Trainium

On Monday, Amazon Web Services released TorchNeuron, a tool that lets AI startups move PyTorch applications onto Amazon’s AI chip, Trainium, by changing one line of configuration. The launch targets the software advantage that helped Nvidia become the default hardware choice for AI applications and fueled its rise as the provider of choice for AI hardware.

Background: How PyTorch And CUDA Became The Default Path

PyTorch is a free, open-source toolkit that developers use to build, train, and run AI models. Facebook originally developed it, and since 2022 it has been part of the Linux Foundation, which maintains open-source projects. Amazon says about 85% of AI apps are written with PyTorch. Nvidia spent years making CUDA, the software infrastructure that allows applications to run on Nvidia GPUs and its other hardware, work smoothly with PyTorch-based apps. That alignment embedded Nvidia hardware in the PyTorch workflow almost by default.

Development: From Porting Convenience To Cloud Economics

TorchNeuron is adapted from PyTorch, so developers who already know PyTorch can use it without a code rewrite. Amazon is not the first company trying to make its chips easier to reach through the toolkit: PyTorch users can also port apps to AMD GPUs or Apple chips with a simple configuration change. Amazon’s argument is that Trainium is not only easy to adopt but also connects AI startups to the company’s entire cloud. AWS presents that as a better startup choice than neoclouds, many of which are tied to Nvidia hardware, including CoreWeave, Nebius, and Firmus.

The Economics AWS Highlights

Colin Brace, Vice President of Technology Software at AWS Annapurna Labs, the unit that makes Amazon’s custom chips, told TechCrunch via email that startups are building on AWS instead of betting on a single-purpose compute provider and scrambling to build everything else around it. He said AWS already provides the infrastructure to go from prototype to production, allowing founders to move fast and scale when needed. AWS also says Trainium handles both training, which teaches the AI model, and inference, which puts the finished model to work when it answers a user’s question. Brace said Trainium’s value is price-performance plus the fact that it is one platform for both training and inference, improving economics across a full workload. Because the chip and the surrounding cloud infrastructure are tightly integrated, AWS argues startups can save money even if it does not offer the cheapest per-token price. Brace pointed to Deepgram, an AI voice application provider and competitor to ElevenLabs, saying it moved to Trainium using native PyTorch support and is processing 70% more audio per dollar at production-grade accuracy.

Latest Progress: Pushing PyTorch Away From Nvidia By Default

The company also said it is working to make PyTorch less automatically centered on Nvidia and will contribute the code for that change to the main PyTorch project. In the blog post announcing TorchNeuron, Amazon wrote that its approach extends beyond its own hardware. The company is contributing changes upstream to PyTorch itself, working with the PyTorch community to remove code that ties the framework to specific chips. The stated goal is to make PyTorch more hardware-agnostic for the entire developer community, not just Trainium users.

Current Stage: Choice, Configuration, And Nvidia’s Next Move

This is the moment where the strategy moves from a simple porting tool to a broader attempt to loosen the tie between PyTorch and Nvidia. For developers, the promise is easier access to Trainium and AWS, plus the possibility of more affordable options. The risk is that PyTorch exists partly to hide the difficult work of making AI code run on hardware, so removing Nvidia-centered defaults may create more configuration headaches. Brace said the real story is removing lock-in that kept developers from having a real choice at all. If Amazon succeeds in deeply uncoupling PyTorch from Nvidia, the consequences for the GPU giant could be larger than the challenge from competing chips alone.