Qualcomm's Modular Platform Aims to Run One AI Software Stack Across GPUs, NPUs and ASICs

On the second day of the 2026 Snapdragon Summit in Hawaii, Chris Lattner, executive vice president of advanced AI software and platforms at Qualcomm Technologies, proposed the Modular platform. The goal is to let the same AI software run across different chip architectures, including GPUs, NPUs and data center ASICs, allowing AI applications to switch more flexibly between on-device and cloud execution.

Lattner said the AI industry's competition is shifting from chips to ecosystems. AI software is often tightly coupled to specific hardware, so developers usually need to re-adapt, optimize and debug when moving to another vendor's chip. In his view, the chip itself is only the foundation; compilers, runtimes, model frameworks, schedulers and toolchains are also key factors for platform adoption.

Qualcomm announced a definitive agreement to acquire Modular on June 24, 2026, and completed the deal on July 29, 2026. Through the acquisition, Qualcomm gained Mojo, a programming language designed for AI and low-level high-performance computing. Mojo has syntax close to Python while aiming for system-level performance and portability.

Microsoft has announced a partnership with Modular to provide native Windows support for Mojo, enabling developers to do system-level programming on Windows with Python-like syntax and build applications for different platforms and vendors. Qualcomm also said Modular's software platform will be added to its Snapdragon portfolio, which spans smartphones, PCs, automotive, wearables and other edge devices. Smooth software integration could accelerate the adoption of on-device AI applications.