Amazon Introduces New AI Chip Tool to Challenge Nvidia’s Dominance

Staff Report: PNN

Amazon has taken a new initiative to reduce Nvidia’s overwhelming influence in the artificial intelligence (AI) chip market. Through its cloud service, Amazon Web Services (AWS), the company has launched a tool called “TorchNeuron.” The tool will allow AI developers to run their applications on Amazon’s own “Trainium” chips instead of Nvidia’s chips without having to rewrite their code.

Unveiling the tool on Monday, Amazon said it would make it easier to transfer applications built using the popular AI development framework PyTorch to Trainium chips. According to the company, developers will not need to make major changes to their code; they will be able to run their applications simply by changing a configuration setting.

Through this initiative, Amazon aims not only to increase the use of its own chips but also to challenge Nvidia’s strong position in the AI software ecosystem.

PyTorch is a widely used open-source software framework for developing, training and running AI models. According to Amazon, around 85 percent of AI applications are currently built on PyTorch. The company believes that making it easier to work with the framework could help increase the adoption of Trainium chips.

PyTorch was originally developed under Facebook’s initiative. Since 2022, it has been operated under the Linux Foundation. Meanwhile, Nvidia has spent years making its “CUDA” software infrastructure compatible with PyTorch. As a result, Nvidia’s graphics processing units, or GPUs, have become the natural choice for many developers running AI applications.

Amazon’s TorchNeuron is an attempt to reduce this dependence. The company says the new tool will be easy to use for developers who are already familiar with PyTorch. However, Amazon is not the only company offering ways to run PyTorch-based applications through configuration changes. Similar options are available for AMD GPUs and Apple chips, potentially intensifying competition in the market.

According to Colin Bres, vice president of technology software at AWS’s Annapurna Labs, which develops Trainium chips, Amazon’s major advantage is that customers can access its chips alongside its entire cloud infrastructure. He said that, rather than relying on a particular computing service provider and subsequently building the necessary infrastructure, AWS already offers the resources required to move from the experimental stage to production.

AWS claims that its Trainium platform can handle both AI model training and inference—the process through which a trained model responds to user queries. The company says that combining these capabilities with its cloud infrastructure could reduce the overall cost of AI operations. However, Amazon is not claiming that its service offers the lowest price in the market on a per-token basis.

To support its claims, AWS has cited the example of Deepgram, an AI-powered voice technology company. According to AWS, after moving to Trainium with the help of PyTorch support, Deepgram can process 70 percent more audio for every dollar spent than it could previously. Amazon also claims that the company has maintained the level of accuracy required for production use. Deepgram competes with ElevenLabs in the AI voice technology market.

Amazon’s plans, however, go beyond simply allowing developers to migrate their applications to Trainium chips. The company is also seeking to introduce changes to the core PyTorch project so that the software framework does not become overly dependent on chips from any particular manufacturer. To this end, Amazon is working to contribute code developed by the company to the main PyTorch project.

If such changes are implemented, competing chipmakers such as AMD and Amazon could benefit alongside Nvidia. Developers could also gain more flexibility in choosing their hardware. However, making software compatible with different chips could also increase configuration complexity.

Nvidia’s strong position in the AI chip market is based not only on its hardware capabilities but also on software support and developers’ long-standing familiarity with its ecosystem. If Amazon succeeds in making PyTorch more hardware-neutral, the competitive landscape could change. However, whether AI developers will genuinely gain access to lower costs and more choices—or instead face new technical complexities—will depend on the tool’s practical performance and its acceptance among developers.