Updated version 1 - Amazon stated that its artificial intelligence chips will support mainstream programming languages, thus challenging Nvidia.

Added the word "shares" in paragraph 2

Stephen Nellis

Reuters San Francisco, October 19 - Amazon.com (AMZN.O) has launched a new tool that enables its in-house AI chips to directly run code written in the most popular AI programming language without any modification, aiming to erode a key advantage behind Nvidia's (NVDA.O) dominance in the AI computing market.

The company's shares rose 5% in early trading, and Morgan Stanley named Amazon as a "top pick."

According to Amazon's chip unit Annapurna Labs, TorchNeuron allows developers to run code written in the language called PyTorch on Amazon's Trainium AI chips with just a single configuration change and no need to rewrite code.

Amazon plans to launch the software at its developer conference in Silicon Valley on Monday and said the software will be released to the public as a free open-source product.

PyTorch, an open-source framework maintained by Meta META.O, is adopted by most AI developers. Nvidia has spent more than a decade developing software tools called Cuda to allow PyTorch code to run efficiently on Nvidia chips.

Other cloud computing vendors, such as Google GOOGL.O, which also has its own AI chips, are working on better PyTorch tools (link) to attract developers away from Nvidia.

Colin Brace, vice president of software at Annapurna Labs, said the move is designed to reduce the number of steps required for AI developers to try Amazon’s own chips.

“This is exactly to meet customers’ current needs,” he said in an interview with Reuters.

Brace said more than 30 early customers—across robotics, world modeling, and speech AI sectors—are already using the new software and that switching to Amazon chips can provide significant cost savings.

(To make news available to non-English speakers, Reuters automatically translates its reports into multiple languages. Since automated translation may contain errors or lack intended context, Reuters does not guarantee the accuracy of the automated translation. It is provided solely for reader convenience. Reuters is not liable for any harm or loss caused by use of the automated translation feature.)