Qualcomm spends $4B on Modular to loosen Nvidia’s AI grip

Qualcomm's near-$4B bet on Modular keeps the AI software startup independent while chasing an open, multi-silicon stack.

ChipNews Staff
2 Min Read

Qualcomm’s nearly $4B acquisition of AI software startup Modular is a bet that open software can pry AI workloads off Nvidia’s exclusive grip. In an interview with EE Times, Modular co-founder Tim Davis said the deal keeps the startup independent while giving its stack global distribution.

Modular’s core products are Mojo, a silicon-agnostic kernel language, and Max, an open modeling and serving framework. Both are open source. Davis argues the future of AI compute is heterogeneous, with inference running across CPUs, ASICs, and edge devices rather than only data-center GPUs. The shift from selling GPU-hours to selling tokens demands software that lets developers ignore the underlying hardware.

Mojo 1.0, a production-ready version after 4.5 years of development, launches shortly. Max reaches general availability in a few weeks, with training support planned later.

The deal has skeptics. One EE Times reader asked how Qualcomm monetizes $4B of open source software. Davis says fungibility is the point: a multi-silicon layer makes alternative hardware easier to finance and deploy, rewiring the economics of cloud compute.

Qualcomm CEO Cristiano Amon and his team, Davis said, are committed to an open, neutral standard that works across all silicon. Whether the bet pays off depends on developers adopting Mojo and Max beyond Modular’s current customers, which include neoclouds, frontier labs, and enterprises.

Share This Article