- Qualcomm is acquiring Modular in an all-stock deal worth approximately $3.92 billion, announced at its investor day on June 24 and expected to close in the second half of 2026.
- Modular was co-founded by Chris Lattner, the engineer who created LLVM and Apple’s Swift programming language: two pieces of infrastructure that most modern software now runs on.
- The deal gives Qualcomm a vendor-neutral AI software platform already used by Oracle, Amazon, Nvidia, and AMD, and doubles Modular’s valuation from $1.6 billion just nine months ago.
Chris Lattner has spent his career building the infrastructure the software world runs on, without most people ever knowing his name. LLVM, the compiler toolkit he created during his PhD at the University of Illinois in the early 2000s, now underpins software development toolchains at Apple, Google, Microsoft, Nvidia, and ARM. Swift, the programming language he started building in secret at Apple in 2010, is what millions of iOS developers write in every day.
Now his latest company, Modular, has been acquired by Qualcomm in an all-stock deal worth approximately $3.92 billion, and this time, the target is the software lock-in that has made Nvidia almost impossible to displace in AI.
The problem Lattner set out to fix
Lattner co-founded Modular in January 2022 alongside Tim Davis, after stints leading AI infrastructure work at Google’s TensorFlow team and heading engineering at RISC-V chip startup SiFive. Their starting point was a diagnosis most AI engineers quietly agree with: the software layer holding the industry together is a mess. Developers building AI applications have to optimise their code separately for each chip architecture a costly duplication of effort that makes the whole ecosystem slower and more expensive than it needs to be.
Modular built a hardware-agnostic platform to fix that. Its flagship products let developers write an AI application once and run it across different processor architectures without rewriting code. The platform does not depend on PyTorch, CUDA, or ROCm, and already runs on chips from Nvidia, AMD, and Apple. In practice: a company running AI inference on Nvidia GPUs in the cloud can shift workloads to Qualcomm chips at the edge without rebuilding its stack.
Modular’s roughly 150-person team now counts Oracle, Amazon, Nvidia, and AMD among its customers.
By September 2025, investors had put $380 million into the company across three rounds, valuing it at $1.6 billion. Backers include GV, General Catalyst, Greylock, DFJ Growth, and Thomas Tull’s US Innovative Technology fund. Today’s acquisition price of $3.92 billion is more than double that figure.
What Qualcomm is buying
Qualcomm has spent decades making the chips inside phones. It is now pushing hard into data centres and edge AI computing, markets Nvidia dominates largely because of CUDA, a software platform that has locked developers into its hardware for nearly 20 years and currently counts over four million developers in its ecosystem. Qualcomm’s chips may be competitive on performance in certain workloads, but without a compelling software story, getting AI teams to look beyond CUDA has been close to impossible.
Modular changes that calculation. Rather than trying to replicate CUDA, Qualcomm is betting on a different strategy: a neutral layer that runs everywhere, including on Nvidia. The pitch to enterprises is to “stop being dependent on any single vendor”
“As agentic AI scales across data centres and edge environments, the industry is moving toward disaggregated, multi-vendor architectures that demand a more open and modern software foundation,” Qualcomm chief executive Cristiano Amon says.
The Modular acquisition also comes as Qualcomm is separately reported to be in advanced talks to acquire AI chip startup Tenstorrent, led by chip architect Jim Keller, at a valuation of between $8 billion and $10 billion, according to The Information. If both deals close, Qualcomm will have spent more than $14 billion assembling an AI infrastructure stack — silicon and software together — that it is betting can compete with Nvidia on terms Nvidia has not faced before.
Lattner has built developer ecosystems that outlasted him twice. The question is whether he can do it a third time, against a company whose software moat has been accumulating for nearly two decades, and in a market where switching costs arecultural.
