Qualcomm announced on 24 June that it would acquire Modular in an all-stock deal worth about $3.9bn, handing the startup’s owners roughly 19.2 million Qualcomm shares. The transaction, expected to close in the second half of 2026, is the clearest sign yet that the mobile-chip company intends to become a serious presence in the data centre, where demand for artificial intelligence hardware has reshaped the semiconductor industry. Modular’s entire workforce of around 150 people will move into Qualcomm’s engineering divisions.
The price tells its own story. Modular, founded in 2022 by Chris Lattner, a well-known compiler engineer, and Tim Davis, raised $250m only nine months ago at a private valuation of $1.6bn. Qualcomm is paying well over double that figure. Such a premium is not a bet on current revenue, which is negligible, but on the strategic scarcity of what Modular has built: a software layer that lets developers run AI models across CPUs, GPUs, NPUs and custom accelerators without rewriting their code for each type of chip.
That capability addresses Qualcomm’s central weakness. The company designs capable processors, but has lacked the software tooling that makes hardware easy to adopt. Nvidia’s dominance of AI computing rests less on its chips alone than on CUDA, the programming platform that locks developers into its ecosystem. Rivals have learned that raw silicon performance means little if engineers must labour to use it. Buying Modular is an attempt to acquire, rather than slowly grow, a credible answer to that advantage.
The deal fits a wider pattern in 2026, in which artificial intelligence, healthcare and energy have driven the bulk of merger activity by value. Within technology, the most contested asset is no longer manufacturing capacity but the software that abstracts away hardware complexity. Acquirers are paying startup-founder prices for small teams whose value lies in specialised expertise and timing rather than in scale. The scarcity of engineers who can write low-level compiler and runtime software has turned a few dozen specialists into strategic assets that chip companies would rather buy than spend years training.
Execution now becomes the test. Qualcomm must integrate Modular’s engineers without dulling the independence that made their work attractive, and persuade developers that its platform is a genuine alternative rather than a defensive purchase. Acquisitions of this kind often falter in the quiet erosion of talent afterwards. For nearly $4bn, Qualcomm has bought time and expertise. Whether it has bought a foothold against Nvidia is the more expensive question.