Qualcomm buys Modular to take on CUDA
Qualcomm is acquiring Modular, maker of the Mojo language and MAX inference engine, in an all-stock deal near $4 billion aimed at Nvidia's CUDA lock-in.
Qualcomm said on June 24 it will acquire Modular, the software company behind the Mojo programming language and the MAX inference engine, in an all-stock deal reporting values at just under $4 billion. The deal should close later in 2026 once regulators sign off. Modular’s co-founders Chris Lattner and Tim Davis, along with a team of roughly 150, join Qualcomm.
The reason this matters is not the price. Modular’s whole pitch is that you write AI code once and run it across very different chips, from Nvidia and AMD GPUs to NPUs and custom ASICs, without rewriting it for each one. That is a direct answer to the thing keeping Nvidia on top: not just its GPUs but CUDA, the software layer most AI code is written against.
Qualcomm CEO Cristiano Amon framed the buy around an industry shift toward “disaggregated, multi-vendor architectures” that need a more open software base. Lattner said joining Qualcomm gives Modular the reach to accelerate that mission.
What it means for you
You will not touch Mojo or MAX directly, and a chip vendor now owning a “neutral” software layer is worth watching with some skepticism. But the direction helps operators. The more AI software runs across many kinds of silicon, the less any single vendor can dictate your prices or lock your stack in place. That is the same force behind OpenAI building its own inference chip and behind cheaper third-party inference clouds. The practical move is unchanged: keep exit ramps in your AI stack so you can move when the economics shift.