Amazon Targets Nvidia’s Hold on PyTorch AI Software

Amazon’s challenge to Nvidia’s influence Amazon is trying to loosen Nvidia’s grip on the software ecosystem used to develop artificial intelligence applications. Its effort centres on PyTorch, a popular framework for building and training AI models. Developers often rely on Nvidia’s chips to run PyTorch workloads, reinforcing the company’s influence beyond the hardware itself. Amazon wants developers to be able to run their PyTorch applications on chips made by other companies, including Amazon’s own. The company is working to make the framework more compatible with alternative processors, reducing the need for developers to adapt or rebuild their applications when they move away from Nvidia hardware. This could give cloud customers more choice over which chips they use. The challenge is not simply to offer different chips, but to make them practical for developers already working with PyTorch. By addressing software compatibility, Amazon aims to make its hardware and other alternatives easier to use within familiar workflows. If developers can run their existing applications across a wider range of chips, Nvidia may face less of an advantage from its close association with the tools developers use. Why PyTorch matters PyTorch is a free software toolkit used to build most AI applications. It gives developers a way to create and train AI models, the systems that learn from data and power many of today’s services. Its importance helps explain why Amazon’s challenge to Nvidia is not just about designing different chips: the tools developers use can also shape which hardware they choose. Nvidia’s influence in AI has been reinforced by CUDA, its software platform for programming its chips. Developers have built up expertise and existing work around that ecosystem, making it difficult to switch to another company’s hardware. Amazon is trying to make its alternative chips more practical for developers by supporting PyTorch, a toolkit many already use. The goal is to let them work with Amazon’s chips without having to abandon a familiar framework. That support matters because hardware alone does not determine whether developers will adopt a chip. If the software they rely on can work with a new platform, trying that platform may become easier. PyTorch therefore sits at the center of Amazon’s effort to make its chips a more credible option for building AI applications. The role of alternative chips Amazon wants more developers to run their PyTorch applications on chips other than Nvidia’s, including Amazon’s own. The effort reflects a broader aim to make it easier for developers to use different hardware without having to leave the tools and workflows they already rely on. PyTorch is widely used to build and train AI models, but applications built with the framework often run on Nvidia chips. Amazon is working to expand support for its own chips, Trainium and Inferentia, so that more PyTorch-based workloads can run on them. It is also seeking to make the framework work across a wider range of hardware, rather than tying developers to one chip maker. That shift matters to Amazon because it could give customers more choice when running AI workloads in the cloud. The company hopes that developers will be able to move applications to Amazon’s chips more easily, while continuing to use PyTorch. In turn, broader compatibility could help Amazon compete for AI computing work that has commonly relied on Nvidia’s hardware. What Amazon’s effort aims to change Amazon’s effort is aimed at changing where developers run AI applications. Rather than relying on Nvidia GPUs in the cloud, developers could use Amazon’s own chips through its cloud services. The company is positioning its Trainium chips for training AI models and its Inferentia chips for running them, while seeking to make them accessible through tools developers already use. A central part of that approach is support for PyTorch, a popular framework for building AI applications. Amazon has worked to make PyTorch compatible with its chips, an effort intended to reduce the need for developers to change how they build applications when choosing different hardware. The goal is to offer another option within cloud computing, where Nvidia has had a strong position. This is an attempt to give developers and businesses a reason to consider running AI workloads on Amazon’s infrastructure and chips. It does not mean Nvidia’s role has already changed: the effort is about making Amazon’s alternative more practical to use. Whether developers adopt that option will depend on how well it meets their needs.