Qualcomm held a separate session at the Snapdragon Summit 2026 conference dedicated to “A unified software layer for AI computing“(Unified Artificial Intelligence Computing Layer). The company intends to create a common set of software tools for artificial intelligence, rather than being tied to equipment from Nvidia or any other manufacturer.
Image source: qualcomm.com
The goal of the project is to eliminate the need to create separate development environments for CPUs, graphics processors, and neural processors. Nvidia has built an ecosystem around its own GPUs and relies on the CUDA platform, while Qualcomm intends to compete and create an open ecosystem covering devices from various manufacturers. Not only NVIDIA and AMD graphics processors are actively entering the AI chip market, but so are neural processors and dedicated AI accelerators from other vendors. But until now, development methods varied by chip type, and moving AI models from one platform to another required extra effort in writing and optimizing code.
The solution to this problem is Modular, a company acquired by Qualcomm for about $4 billion. The key components of its ecosystem are the Mojo programming language, the MAX platform for running AI models directly on computing devices, and the Modular Cloud cloud service for managing large-scale AI infrastructure. According to Qualcomm, the modular software environment currently supports six architectures from six major hardware manufacturers.
«Modular’s main mission is to make heterogeneous computing open“Chris Lattner, co-founder of Modular and former senior vice president of advanced software and platforms at Qualcomm, said. In his view, developers should be able to choose the most appropriate device for a specific task without having to be tied to a single vendor.
Over the years, Nvidia has accumulated a rich library of software tools, and the CUDA platform has become its strong competitive advantage. Qualcomm is committed to expanding choice by providing a single development environment that does not change across hardware platforms. Starting with smartphone processors, Qualcomm has expanded into personal computers, automobiles, wearable devices and data center artificial intelligence accelerators. The broader the product range, the more important software can integrate different chips into a common system. Even before the official takeover, Qualcomm and Modular were working on implementing the latter’s software solutions into the Dragonwing AI accelerator in the data center. The deal has been completed and the work will be moved to the Snapdragon platform for PCs and smartphones – the porting process has already begun for Modular products including Mojo and MAX. The unified development environment will help you easily transfer the AI models built for the data center to PCs and smartphones.
A partnership with Microsoft is possible – together with the software giant, efforts are underway to provide support for the Mojo language in the Windows environment without the need for additional transcoding programs. This will help attract many Windows application developers to join the modular ecosystem. The focus of the AI industry is shifting from prioritizing hardware performance to providing an ecosystem for developers. Nvidia tied its accelerators strictly to the CUDA platform, while Qualcomm decided to consolidate all its solutions into a common software platform. But this process is still in its early stages.
If you find an error, select it with your mouse and press CTRL+ENTER.
