Amazon said on 19 October 2026 that TorchNeuron, its new software for AWS Trainium chips, “runs existing PyTorch code on Trainium with no rewrite”, and that it will be released under the Apache 2.0 open-source licence at PyTorch Conference North America on 20 October. Reuters reported the same day that the tool is Amazon “seeking to erode the advantage that underpins Nvidia’s dominance in the AI computing market.”
Amazon says “85% of AI researchers and developers write their code in PyTorch.” Colin Brace, vice president of software at Amazon’s Annapurna Labs chip unit, told Reuters that more than 30 early customers are already using the software.
Business Pill · Switching Costs
A short explainer on switching costs. It teaches the general idea only and says nothing about any company in this story.
The key insight: As we read it, Amazon is attacking the cost of switching rather than the price of the chip: its post puts engineering time ahead of hardware as the biggest expense for many startups, and TorchNeuron is built to remove the rewrite.
Amazon’s post says that with TorchNeuron, “the PyTorch code developers already write runs natively on Amazon’s AWS Trainium chips, with a single configuration change and no rewrite.” It supports Trainium2 and Trainium3.
The post explains the mechanism: “PyTorch already has a built-in way for new hardware to register itself. TorchNeuron plugs into that standard mechanism, meaning the tools developers already use to test, debug, and scale their AI models all work on Trainium without modification.”
In a demonstration, Amazon says its engineers ran the same training script on “a conventional chip” and on Trainium, and that “the AI model learned the same things at the same rate.” Amazon says it tests TorchNeuron “against hundreds of thousands of AI models on every build.”
How PyTorch code reaches Amazon’s Trainium chips with TorchNeuron, as Amazon describes it on 19 October 2026: existing PyTorch code, one configuration change, the TorchNeuron backend (open-sourced under Apache 2.0 on 20 October) and Trainium2 or Trainium3. The 85% share is Amazon’s; the more than 30 early customers are Colin Brace’s figure as reported by Reuters.
Amazon’s own post at 13:57 UTC on 19 October, an X article titled “Amazon opens its AWS Trainium AI chips to every PyTorch developer, no rewrite needed”. It carries the figures this piece quotes from Amazon, including that 85% of AI researchers and developers write in PyTorch and the 20 October open-source release, and it says “startups and enterprises in robotics, video, world models, self-driving cars, speech, drug discovery, LLMs, and finance are already building on it.”
AMAZON: AI CHIPS WILL WORK WITH POPULAR CODING LANGUAGE, CHALLENGING NVIDIA
First Squawk’s headline post at 13:05 UTC on 19 October, about five minutes after Amazon’s post went up, carrying the Nvidia framing that Reuters’ report uses. Amazon’s own post does not name Nvidia; the comparison in this piece is attributed to Reuters.
AMAZON: TORCHNEURON TO BE OPEN-SOURCED AT PYTORCH CONFERENCE ON OCTOBER 20
A second First Squawk post a few seconds later, on the open-source release at PyTorch Conference on 20 October. It matches the date in Amazon’s post, which adds that the licence is Apache 2.0.
The CUDA Question
Amazon’s post does not name Nvidia. Reuters does. It reports that PyTorch is “used by most AI developers” and that “Nvidia has spent more than a decade on its software tool called Cuda that makes PyTorch code run well on Nvidia chips.”
Reuters adds that “Other cloud computing players such as Google, which also has its own custom AI chips, are also working to lure away developers from Nvidia with better PyTorch tools.”
Amazon frames the move as choice: “When developers can move between chips without penalty, chipmakers have to earn developers’ business, and developers benefit from lower costs and faster iteration.”
TorchNeuron by the numbers, from Amazon’s post and Reuters’ report of 19 October 2026. The 30+ early customers are Colin Brace’s figure as reported by Reuters; every other figure is Amazon’s.
Who Is Using It
Amazon names two early adopters. It says Cartesia “ran its standard inference pipeline with minimal changes to its PyTorch code”, and that TorchNeuron “allows the Lieber Institute to explore these strategies on Trainium” in drug discovery work.
Brace told Reuters the early customers span “robotics, world models and voice AI”. Reuters reports he said that the switch to Amazon’s chips “can result in steep cost savings.” His stated aim, per Reuters, is reducing the number of steps AI developers must take to try Amazon’s chips: “This has been about meeting customers where they’re at.”
The Wider Toolchain
Amazon says it is “contributing changes upstream to PyTorch itself”, working with the community “to remove code that ties the framework to specific chips.” It says Hugging Face “has enabled native Trainium support directly in its popular AI development tools.”
TorchNeuron also “integrates with TorchTitan, PyTorch’s native distributed training framework”, according to the post. Amazon says it has made “a major open-source contribution to vLLM” and “open-sourced its complete set of performance kernels.”
The Cost Argument
Amazon’s case is about engineering time as much as chip prices. Its post says: “For many companies, especially startups, the most expensive line item is engineering time.” It adds that large enterprises can shift capacity “that served inference workloads” to training jobs during off-peak hours.
AWS’s own page for its Trainium2-based instances says “Trn2 instances offer 30-40% better priceperformance than GPU-based EC2 P5e and P5en instances.” That is AWS’s claim; neither Amazon’s post nor Reuters gives a price or performance figure for TorchNeuron itself.
The Structural Read
As we read it, the software is free and the hardware is where Amazon earns: TorchNeuron is released under Apache 2.0, while Trainium capacity is sold as AWS instances, which AWS says offer “30-40% better priceperformance than GPU-based EC2 P5e and P5en instances.”
As we read it, the target is a habit more than a chip: Reuters reports Nvidia “has spent more than a decade” on Cuda, and Amazon’s answer is to meet developers in PyTorch, the framework its post says 85% of them already use.
As we read it, the proof will come from named workloads: Amazon names Cartesia and the Lieber Institute, Reuters counts more than 30 early customers, and neither source gives a benchmark against Nvidia hardware for TorchNeuron.
Colin Brace, vice president of software, Annapurna Labs, to Reuters, 19 October 2026
“This has been about meeting customers where they’re at”
Three Implications
SWITCHING COSTS FALLAmazon says a single configuration change moves PyTorch code to Trainium, and that standard PyTorch tools for testing, debugging and scaling work without modification.
AN OPEN RELEASEAmazon says TorchNeuron will be released under the Apache 2.0 licence at PyTorch Conference North America on 20 October; Reuters calls it a free and open-source offering.
THE COST CLAIM IS AWS’S AWS says its Trn2 instances offer 30-40% better priceperformance than its GPU-based P5e and P5en instances; Reuters reports Brace said the switch can bring steep cost savings, without a figure.
The framework puts it this way: “Close the SCARCE layer (keep proprietary) + Open the ABUNDANT layer (commoditize).”
As we read it, TorchNeuron opens the software layer under Apache 2.0 while Amazon keeps the Trainium chips, its own silicon, as the layer it rents out. The framework was published in March 2026 and does not discuss Amazon, Nvidia or TorchNeuron.
What Is Not Established
Neither source quantifies the “steep cost savings” Brace described, and Amazon names only two of the more than 30 early customers Reuters counts. The two sources also describe the timing differently: Reuters says Amazon plans to announce the software at a developer conference on Monday, while Amazon says the open-source release comes at PyTorch Conference North America on 20 October. How closely TorchNeuron matches CUDA’s performance on the same models is not stated in either source.
Business Engineer Framework
The Open vs Closed Meta-Framework
A Business Engineer framework on which layer of a technology stack to open and which to keep closed.
Amazon said on 19 October that TorchNeuron lets existing PyTorch code run on its Trainium2 and Trainium3 chips with one configuration change, and that the software will be open-sourced under Apache 2.0 on 20 October. Reuters reports more than 30 early customers and describes the move as an attempt to erode the software advantage behind Nvidia’s position.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. This piece is based on Amazon’s own post of 19 October 2026 and Reuters’ report of the same day (read via its StreetInsider syndication), both read in full, and on AWS’s page for its Trn2 instances. The Nvidia and CUDA framing is Reuters’; Amazon’s post does not name Nvidia. The 30-40% price-performance figure is AWS’s own claim. Nothing here is a forecast, and nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
