The tool, called TorchNeuron, will allow developers to run code written in PyTorch, one of the world’s most widely used AI programming frameworks, on Amazon’s Trainium chips without rewriting their existing code. According to Amazon’s Annapurna Labs chip unit, developers will need to make only a single configuration change to run their applications on the company’s hardware.
Amazon plans to unveil the software at a developer conference in Silicon Valley on Monday and release it publicly as a free, open-source offering, a move that could help lower barriers for businesses seeking alternatives to Nvidia’s AI computing technology.
PyTorch, an open-source framework stewarded by Meta, is widely used by AI developers to build and train artificial intelligence models. Nvidia has maintained a strong position in the market partly through its CUDA software platform, which it has spent more than a decade developing and refining to enable PyTorch-based applications to run efficiently on its chips.
By making it easier to transfer existing AI workloads to its own processors, Amazon hopes to weaken one of the key advantages that has helped Nvidia maintain its leading position in the industry.
Colin Brace, vice president of software at Annapurna Labs, said the initiative was intended to simplify the process of testing and adopting Amazon’s in-house chips.
“This has been about meeting customers where they're at,” Brace told Reuters in an interview.
Brace said more than 30 early customers across fields including robotics, world models and voice AI were already using the new software. He added that moving workloads to Amazon’s chips could deliver substantial cost savings, potentially making the company’s hardware more attractive to organisations facing rising demand for AI computing resources.
Amazon is not alone in seeking to reduce developers’ dependence on Nvidia. Other major cloud computing companies, including Google, have also developed their own AI chips and are working to attract customers by improving support for PyTorch and making their hardware easier to adopt.
The competition reflects a broader struggle in the AI industry, where access to powerful computing hardware and the software needed to run it efficiently has become a critical advantage. While alternative processors may offer cost benefits, developers’ reliance on established software tools remains a major factor in their choice of hardware.
With TorchNeuron, Amazon is seeking to address that challenge directly by allowing developers to retain their existing code rather than undertake costly and time-consuming rewrites. Whether the approach will persuade more AI developers to move away from Nvidia will depend on the software’s performance, reliability and the savings it can deliver in practical use.