.oesnadaWhere have you been?

Developing story

Amazon and Google PyTorch Tools Challenging Nvidia

Amazon and Google PyTorch Tools Challenging Nvidia: 1 development over 1 day, since 20 October 2026. Latest, 20 October 2026: Google announced Tuesday that TorchTPU, its version of the PyTorch AI development toolkit, is now available as an open-source project.

Oesnada automated editorial system · Last updated: (2026-10-20)

The story so far

Amazon released TorchNeuron on 19 October, a software tool that lets code written in PyTorch run on its Trainium AI chips after a basic configuration change 1. PyTorch is the open-source framework many AI developers use 1. The tool had been in private beta since late last year, and Amazon said it was now open to all developers as a public beta 1.

Amazon Web Services described the release as a way to help AI startups move their applications onto Trainium, its custom AI chip 2. The tool works with PyTorch, a free, open-source toolkit for building, training and running AI models 2.

On 20 October, Google announced that TorchTPU, its version of the PyTorch AI development toolkit, was now available as an open-source project 3. The announcement followed Amazon's release and, like it, targets Nvidia's dominant position among AI developers 3.

Written automatically from the Oesnada stories in this timeline. Every number is checked against them. The numbers in brackets link to those stories.

Timeline

  1. TechCrunch

    Google Joins Amazon in Challenging Nvidia's Hold on AI App Developers

    Google announced Tuesday that TorchTPU, its version of the PyTorch AI development toolkit, is now available as an open-source project.

  2. Also on our radarAmazon Launches TorchNeuron Dev Tool in Its Biggest Challenge to Nvidia YetTechcrunch

    Amazon Web Services released TorchNeuron on Monday, a tool meant to let AI startups move their applications onto Trainium, Amazon's custom AI chip. TorchNeuron works with PyTorch, a free, open-source toolkit developers use to build, train, and run AI models. According to Amazon, apps written in PyTorch can run on Trainium by changing a single line of configuration, with no code rewrite. The move targets the software advantage that has made Nvidia the default hardware choice for AI. Nvidia has spent years making its CUDA software, which lets apps run on its GPUs, work smoothly with PyTorch, which Amazon says underpins about 85% of AI apps. PyTorch users can already shift to AMD GPUs or Apple chips with a simple configuration change, but AWS argues Trainium also opens access to Amazon's full cloud. It pitches that as a better option than neoclouds, such as CoreWeave, Nebius, and Firmus, many of which are tied to Nvidia hardware. Colin Brace, VP of Technology Software at AWS Annapurna Labs, the unit that builds Amazon's custom chips, said startups on AWS aren't betting on a single-purpose compute provider. AWS says Trainium handles both training and inference, meaning teaching a model and then running it to answer user queries, which it argues improves costs across a full workload even if AWS isn't the cheapest per token. Brace pointed to voice AI company Deepgram, which he said processes 70% more audio per dollar on Trainium at production-grade accuracy. Amazon also says it is contributing code to the main PyTorch project to remove ties to specific chips and make the framework more hardware-agnostic. Whether that gives developers more choice and lower costs, as Amazon promises, or just more configuration work remains to be seen.

    Read at the source
  3. Also on our radarAmazon Launches TorchNeuron to Run PyTorch Code on Trainium Chips WithoutGeekwire

    Amazon has released TorchNeuron, a software tool that lets code written in PyTorch, the open-source framework many AI developers use, run on Amazon's Trainium AI chips after a basic configuration change. The tool has been in private beta since late last year, and Amazon said Monday it is now open to all developers as a public beta. Amazon plans to release it as open-source software on Tuesday, alongside the PyTorch Conference in San Jose, California. Until now, developers had to rewrite PyTorch code to use programming interfaces specific to Amazon's chips, said Colin Brace, vice president of technology software at Annapurna Labs, Amazon's chip-design unit. "If you know PyTorch, then you know Trainium," he said. The tool targets one of Nvidia's biggest advantages. Nvidia dominates the AI chip market partly because much of the industry's software was built around its CUDA platform, and adapting that code to other hardware can take months of engineering. That has kept many companies from switching, even as Amazon, Google and others build their own chips. Google is working on a similar project, TorchTPU, to run PyTorch natively on its TPU chips, and Microsoft said last week it plans to run most of its AI services on its own Maia chips. Amazon is also reportedly in talks to acquire Decart, a startup whose software helps move AI models from Nvidia chips to alternatives such as Trainium, in a deal that could value it at about $7 billion. More than 30 companies and internal Amazon teams have used TorchNeuron during the private beta. Deepgram, a speech AI startup, says its benchmarks showed Trainium processing about 60% more audio per dollar than its previous GPU setup, and it now runs Trainium in production. Amazon's Prime Video team used the tool to build Prime 360, an AI system that reconstructs NBA plays in 3D, which will appear on select broadcasts this season. Amazon says it tests TorchNeuron nightly against more than 100,000 models from Hugging Face. Brace said newer PyTorch versions can detect the hardware they run on, so some developers may not need to change their code at all. Trainium's biggest users so far are Anthropic and OpenAI, both closely tied to Amazon financially.

    Read at the source

Methodology and sources

Oesnada groups published stories into a developing story when a language model judges that they continue the same thread. The grouping is automatic and can be wrong. A thread gets a page once it has at least 3 developments spread over at least 2 different days. The one-line updates and the summary are written automatically from Oesnada stories only, and every number in them is checked against those stories.

Cite this page

Oesnada, "Amazon and Google PyTorch Tools Challenging Nvidia", https://oesnada.com/storylines/amazon-and-google-pytorch-tools-challenging-nvidia, accessed 21 October 2026.