Qualcomm has closed a deal to acquire Modular, a three-year-old startup, for nearly $4 billion. The entire team of roughly 150 people, including both founders, will join Qualcomm. The deal is expected to close in the second half of this year.
To understand why Qualcomm paid that much, you need to understand the problem Modular was solving.
Right now, Nvidia controls somewhere between 75 and 90 percent of the market for the specialized chips that run AI workloads. That dominance is not just about the chips themselves. Nvidia built a software system called CUDA over nearly 20 years, and almost all AI software in the world is written to run on it. Over 4 million developers have registered for CUDA, and more than 40,000 organizations use applications built on top of it. That is a deep, structural lock-in: if you want to switch to a cheaper competitor's chip, your engineers face months of rewriting code.
Modular attacked that problem directly. Its software, including a programming language called Mojo and a tool called MAX, lets developers write AI code once and run it on chips from Nvidia, AMD, Intel, and Qualcomm without costly rewrites. Modular claimed this approach could reduce costs by up to 80 percent compared to running on vendor-specific systems. Whether those numbers hold up in every situation, the core idea, a chip-agnostic software layer, is genuinely valuable to any chipmaker trying to pull customers away from Nvidia.
That is exactly Qualcomm's position. The company makes the chips inside most of the world's smartphones, but that market is maturing. CEO Cristiano Amon has been explicit about wanting to push into data centers, the large computer facilities that run AI services for businesses globally. Qualcomm actually exited the data center market in 2018 to focus on smartphones, and is now fighting its way back in.
The Modular acquisition is one piece of a larger buying spree. Qualcomm already acquired Ventana Micro Systems, a startup building server chips, and Alphawave Semi for about $2.4 billion to improve high-speed data connections. It is also in separate talks to acquire Tenstorrent, an AI chip startup, for up to $10 billion. Add those numbers together and Qualcomm could spend close to $14 billion on AI acquisitions within a few months.
The logic is clear: Qualcomm needs the hardware, the connectivity plumbing, and now the software layer that makes its chips attractive to businesses already comfortable with Nvidia. Modular specifically is the piece that reduces the switching cost. If Qualcomm's chips are as good as Nvidia's on price or performance, but switching requires months of expensive engineering work, most companies will not bother. Modular's tools remove that excuse.
For business operators, the practical takeaway is this: competition in the AI chip market is increasing, and software like Modular's is the reason it can increase at all. More competition tends to push down prices and improve service quality. Companies currently paying high prices for Nvidia-based cloud AI services, or locked into contracts with a single cloud provider, should watch this space. The barriers to switching providers are starting to come down, and that gives buyers more options than they had a year ago.
What remains an open question is whether Qualcomm can actually execute. Buying startups is straightforward. Keeping their talent, shipping competitive products, and convincing large enterprises to trust a new entrant in a market Nvidia has owned for years: that is the hard part.