Qualcomm outlined a plan to use newly acquired software company Modular to build a unified AI development platform that can run across multiple chip types and compete with Nvidia’s ecosystem.. The company completed its acquisition of Modular in July after announcing the all-stock deal on June 24, valued at about $3.92 billion, according to Reuters..
The acquisition brought Chris Lattner, creator of LLVM and the Swift programming language, to Qualcomm as senior vice president for advanced AI software and platforms.. At a dedicated session on Sept. 23 at Snapdragon Summit 2026, Lattner said Modular’s technology is designed to address a problem for AI developers who now have to write and optimize code separately for GPUs, NPUs and dedicated AI accelerators..
Modular’s software stack includes the Mojo programming language, the MAX inference engine and Modular Cloud for deployment. The stack currently supports six chip architectures from six hardware vendors in one software environment.. Lattner described the effort as a direct alternative to Nvidia’s CUDA platform, which has become a standard software layer for Nvidia GPUs..
“When we run on an Nvidia chip, we replace all of CUDA,” Lattner told EE Times.. “Modular’s core mission is to open up heterogeneous compute,” Lattner said at the summit. “Developers should be able to choose the hardware that best fits the purpose without being tied to specific hardware.”.
Qualcomm said it plans to extend the Modular software stack beyond its Dragonwing data center AI accelerator to Snapdragon chips for PCs and smartphones.. The company has also partnered with Microsoft to support Mojo on Windows without a separate conversion process.. “We are bringing the entire Modular software platform across Qualcomm’s full portfolio, and that includes Snapdragon,” Lattner said..
Featured image credit. Tags: ModularQualcomm
