- Qualcomm plans to acquire Modular in an all-stock transaction valued at approximately $3.92 billion, announced at an investor day on June 24 and expected to close in the second half of 2026.
- Modular was co-founded by Chris Lattner, an engineer who created LLVM and Apple’s Swift programming language, two infrastructures on which most modern software runs.
- 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 that the software world runs on, without most people knowing his name. LLVM is a compiler toolkit he created as a doctoral student at the University of Illinois in the early 2000s and now powers software development toolchains at Apple, Google, Microsoft, Nvidia, and ARM. Swift, the programming language he started developing in secret at Apple in 2010, is the language millions of iOS developers write every day.
Now, his latest company, Modular, has been acquired by Qualcomm in an all-stock deal worth about $3.92 billion. This time, the target is the software lock-in that has made it nearly impossible to replace Nvidia in the AI space.
The problem Ratner tried to solve
Ratner co-founded Modular with Tim Davis in January 2022 after leading the AI infrastructure practice on Google’s TensorFlow team and serving as head of engineering at RISC-V chip startup SiFive. Their starting point was a diagnosis that most AI engineers quietly agreed with. It was that the software layer that held the industry together was in disarray. Developers building AI applications must optimize code separately for each chip architecture, making the entire ecosystem unnecessarily slow and expensive due to costly duplication of effort.
Modular has built a hardware-agnostic platform to solve that problem. The company’s flagship product allows developers to write AI applications once and run them across different processor architectures without rewriting code. The platform is PyTorch, CUDA, and ROCm independent and already runs on chips from Nvidia, AMD, and Apple. In fact, companies running AI inference on Nvidia GPUs in the cloud can move their workloads to Qualcomm chips at the edge without rebuilding their stacks.
Modular’s team of approximately 150 people currently counts Oracle, Amazon, Nvidia, and AMD among its customers.
By September 2025, investors had invested $380 million in three rounds, giving the company a valuation of $1.6 billion. Supporters include GV, General Catalyst, Greylock, DFJ Growth, and Thomas Tull’s US Innovative Technology Fund. Today’s acquisition price is $3.92 billion, more than double that amount.
What Qualcomm will acquire
Qualcomm has spent decades making the chips that go into cell phones. The company is currently focused on data center and edge AI computing, a market that NVIDIA dominates thanks in large part to its software platform CUDA, which has locked developers to the company’s hardware for nearly two decades and now has more than 4 million developers in its ecosystem. Qualcomm’s chips may be competitive in performance for certain workloads, but without a compelling software story, it’s nearly impossible to convince AI teams to look beyond CUDA.
Modular changes that calculation. Rather than trying to replicate CUDA, Qualcomm is betting on a different strategy: a neutral tier that runs everywhere, including Nvidia. The pitch to companies is to “stop relying on a single vendor.”
“As agent AI extends across data centers and edge environments, the industry is moving toward a fragmented, multivendor architecture that requires a more open and modern software foundation,” said Cristiano Amon, CEO of Qualcomm.
Modular’s acquisition comes amid separate reports that Qualcomm is in talks to acquire AI chip startup Tenstorrent, led by chip architect Jim Keller, at a valuation of $8 billion to $10 billion, according to The Information. If both deals go through, Qualcomm is betting it will spend more than $14 billion building an AI infrastructure stack (combining silicon and software) that will allow it to compete with NVIDIA on conditions the company has never faced before.
Ratner built a developer ecosystem that surpassed him twice. The question is whether he can make a third try, against a company with a software moat built up over nearly two decades, and in a market where switching costs are cultural.
