Qualcomm to buy Modular for $3.92 billion in all-stock bet against CUDA
The chipmaker's biggest software acquisition yet brings in Chris Lattner's vendor-neutral AI inference stack as Qualcomm builds out a data center business.
Qualcomm said Tuesday it’ll acquire Modular Inc. in an all-stock deal valued at roughly $3.92 billion, the chipmaker’s largest software purchase to date and a direct wager that the next phase of AI infrastructure will be won at the compiler layer, not the silicon layer.
The transaction, announced at Qualcomm’s New York investor day on June 24, will issue up to 19.2 million new Qualcomm shares to Modular’s equity holders through a private placement, according to the Form 8-K filed this week. Reuters and the Wall Street Journal pegged the deal at $3.92 billion based on Qualcomm’s Tuesday close of $204.13; Bloomberg, which first reported advanced talks on June 22, had put the number closer to $4 billion. The definitive agreement was signed June 21, with closing expected in the second half of 2026.
For Modular, it’s a sharp markup. The company was valued at $1.6 billion in a September 2025 round and had raised roughly $380 million in total since being founded in 2022 by Chris Lattner, the creator of LLVM and Swift at Apple, and Tim Davis. Around 150 employees are expected to move to Qualcomm. Its core assets are the Mojo programming language and the MAX inference platform, both built to run AI workloads across heterogeneous hardware without the developer pulled into a single vendor’s stack.
That’s the entire strategic point. Nvidia’s CUDA has functioned for more than a decade as the industry’s gravitational center, locking developer habits to Nvidia silicon in a way no competing accelerator has cracked. Buying Modular is Qualcomm’s attempt to acquire, rather than rebuild, the vendor-neutral software layer that would make alternative hardware actually shippable.
CEO Cristiano Amon told investors the acquisition “deepens the software foundation for Qualcomm Technologies’ data center strategy,” and argued that “the future belongs to developer-friendly, horizontal platforms that can run across diverse compute environments and give customers real choice in how and where they deploy AI.” On the same stage, Qualcomm unveiled the Dragonfly C1000 server CPU, naming Meta as its first hyperscaler customer.
The framing is familiar from prior platform inflections, the post-2008 unbundling of financial infrastructure or the early-2010s mobile OS wars: incumbents don’t lose by being out-engineered, they lose when a horizontal layer makes their moat optional. Qualcomm is paying $3.92 billion to be the company offering that layer, rather than the one defending against it.