Modular has open-sourced its AI software platform, aiming to provide a common framework for running inference across hardware from Qualcomm, AMD, Nvidia and other accelerator platforms.
Modular has open-sourced its software platform at ModCon 2026, making the code available under the Apache 2.0 licence with LLVM exceptions. The software stack is designed to run AI inference across different computing platforms, including Qualcomm Snapdragon systems, AMD platforms, Nvidia GPUs, AWS Trainium, Google TPUs and Qualcomm’s AI accelerators.
The platform aims to reduce the complexity of adapting AI software for different hardware systems. Instead of requiring extensive changes when moving an application between accelerator platforms, Modular provides a common software framework for developing and deploying AI workloads across different computing systems.
AMD’s involvement in the development of the Modular software stack was another highlight of the conference. The platform is designed to support a range of hardware, including Apple Mac Mini, Nvidia DGX Spark, AMD Ryzen AI Halo and Qualcomm Snapdragon platforms, as well as larger accelerator systems.
The open-source release gives developers access to Modular’s codebase, allowing them to examine and contribute to the software stack. The approach is intended to reduce the software development work required when supporting new AI hardware while maintaining a common development environment across different processors and accelerators.
The release follows Qualcomm’s acquisition of Modular and is part of its effort to develop an AI software platform that can work across different hardware vendors. With the code now available under an open-source licence and support spanning multiple accelerator architectures, Modular is positioning its platform as a common software layer for AI development and inference.
