MAUI, Hawaii - Qualcomm is building an integrated software environment to develop and optimise AI across different hardware, led by Modular, which it acquired. The move targets the burden on developers of having to tailor AI workloads for each accelerator and architecture as devices and computing environments diversify. Qualcomm CEO Cristiano Amon (크리스티아노 아몬) explained the rationale for the Modular acquisition in a keynote at Snapdragon Summit 2026 in Maui, Hawaii, on Sept. 22.

Qualcomm pointed to the optimisation burden in heterogeneous computing environments. AI workloads currently run on different accelerators, hardware architectures and platforms. Qualcomm said developers must optimise workloads separately for each environment to deliver a consistent AI experience across multiple devices.

Snapdragon also has multiple engines that handle AI computing. Qualcomm introduced the Adreno Matrix Core in the next-generation mobile platform unveiled at the summit, expanding matrix acceleration from the NPU and CPU to the GPU. This is seen as a structure in which software plays a growing role in deciding how to allocate workloads among engines even within a single platform.

Qualcomm plans to use Modular to provide an integrated software environment for developing and optimising AI across heterogeneous computing environments. Qualcomm said Modular serves as an open software layer that supports moving and running AI workloads across different computing environments. It aims to let developers deploy AI in a single software environment without repeating work for each hardware type.

The hardware supported by Modular is not limited to Qualcomm products. Qualcomm said the Modular platform aims for a hardware-agnostic structure that supports CPUs, GPUs and NPUs developed by multiple companies, as well as custom silicon. The goal is a heterogeneous computing environment not tied to a single hardware, software or cloud provider.

Qualcomm emphasised that it has AI hardware, connectivity and deployment capabilities spanning the cloud and data centres, edge and devices. The move by a chip supplier to secure software that supports competitors' hardware is seen as an effort to expand its position in the AI software layer.

Amon said, "We will open-source Modular's Mojo and MAX to support running AI workloads not only on Snapdragon but also on various hardware architectures." He added, "Our long-term goal is to build an open AI software platform that the entire industry can use."

Open-sourcing Mojo and MAX aims to build an industry-shared AI software platform.

Qualcomm also linked the Modular acquisition to its edge AI business. It said combining Qualcomm's edge AI technology with Modular's open software platform could broaden AI investment concentrated in centralised infrastructure to devices, cars, robotics and industrial sites.

It also makes it possible to raise utilisation of domestic and overseas computing infrastructure and shift from centralised AI to distributed, interoperable AI infrastructure. Qualcomm said it will support running not only large frontier models but also national large language models, industry-specific models and open-weight models on multiple AI hardware systems.

Qualcomm also tied this structure to South Korea's sovereign AI strategy. Qualcomm said an open strategy allows the government and domestic institutions to retain control over data, models and digital infrastructure while choosing infrastructure, cloud, AI models and hardware platforms as needed.

This would provide a foundation for domestic developers, companies and research institutions to develop and deploy their own hardware and software without being dependent on a specific vendor. Qualcomm added that this would support South Korea's AI G3 goal and broaden opportunities for AI hardware and software developed in the country to participate in the global ecosystem.