SAN DIEGO — Qualcomm is in advanced talks to acquire Modular, an artificial intelligence chip startup whose software lets developers build and deploy AI models without learning the low-level details of each processor, according to people familiar with the matter. A deal would value the startup at roughly $4 billion, the people said, though the talks have not produced a final agreement and could still fall apart.
Modular was founded in 2022 by Chris Lattner, the computer scientist who created the Swift programming language at Apple and later ran Google’s TensorFlow team, and Tim Davis, a former Google engineer. The company built its reputation on two things: Mojo, a programming language designed for AI that runs close to the speed of hand-optimized code, and MAX, a platform that lets the same AI models run across different kinds of hardware.
The potential acquisition fits Qualcomm’s strategy of extending its chips into the data center and the AI PC market. The company’s Snapdragon processors have a strong position in phones, and its push into laptop processors has made headway with Microsoft’s Copilot+ PC initiative, but its data center ambitions have been limited. Modular’s software would give Qualcomm a way to make its hardware easier for developers to adopt, a bottleneck that has slowed the company’s entry into markets where Nvidia’s software ecosystem dominates.
The price tag, roughly $4 billion, reflects both the startup’s promise and the froth of the current AI market. Modular has raised about $100 million from investors including General Catalyst, and its revenue is believed to be small, according to people familiar with its finances. The premium, the people said, is for the team and the software platform rather than for existing sales, a calculation that is becoming common in AI acquisitions as companies pay for capability they cannot build quickly themselves.
Qualcomm has a history of buying its way into new markets. The company paid $1.4 billion for Nuvia in 2021, acquiring a team of chip architects who had left Apple and Arm, and the Nuvia designs became the foundation of Qualcomm’s Snapdragon X laptop processors. The Modular talks follow the same playbook: buy the team and the technology, integrate them into a larger product line, and let the distribution engine of a large company carry the result.
The AI chip market has been consolidating at a rapid pace. Nvidia has expanded through investments and acquisitions, AMD has absorbed startups to build its software stack, and cloud providers have begun designing their own silicon in-house. For a company the size of Qualcomm, the choice is increasingly between buying capability and being left out of the fastest-growing segment of the chip market, analysts said.
The talks also reflect a shift in how AI hardware is evaluated. For years, the industry competed on raw processing power, measured in teraflops and memory bandwidth, but the bottleneck for most developers is software: writing AI code that runs efficiently on unfamiliar hardware is difficult, and the winner in each market has usually been the company that made its platform easiest to use. Modular’s software addresses that bottleneck directly, which is why several companies have reportedly studied the startup before Qualcomm’s current talks.
There are obstacles between the talks and a completed deal. The valuation has been a point of contention, according to people familiar with the discussions, and regulatory scrutiny of AI acquisitions has increased in both the United States and Europe. A deal of this size would not trigger the thresholds that attract the most aggressive antitrust review, but the review process would add months to any closing, the people said.
The outcome matters beyond the two companies. If Qualcomm closes the deal, it will have acquired one of the most respected software teams in AI, and the integration will test whether a chip company can operate a developer platform successfully. If the talks collapse, Modular will remain an independent company with a rare asset: software talent that the largest companies in the industry have shown they want to own.
The deal would also give Qualcomm a position in the software layer of AI, a market that has been dominated by Nvidia’s CUDA platform. Nvidia’s success has rested as much on its software ecosystem as on its hardware, and competitors have struggled to match the ease with which developers can write code that runs on Nvidia chips. Modular’s founders built Mojo with exactly that problem in mind, and the language has attracted a following among developers who want the performance of low-level code without the difficulty of writing it. An acquisition would not immediately displace CUDA, analysts said, but it would give Qualcomm a credible answer to the question its sales force has faced for years: why should developers bother writing code for chips that are not Nvidia’s?
The negotiations are the latest sign that AI consolidation is moving beyond the largest companies. The market has seen a wave of deals this year, with chip makers and cloud providers acquiring startups whose technology they cannot build quickly, . Modular’s founders have not said whether they would accept an acquisition or continue building independently. The talks could still collapse, but the fact that Qualcomm is willing to pay $4 billion for a software company with modest revenue is the clearest evidence yet of how the AI industry values the people who make hardware usable.
For Qualcomm’s shareholders, the acquisition would be a bet on a simple thesis: that the next phase of AI growth belongs to chips that are easy to program, and that the company which controls the software layer will capture more of the value than the company which merely sells the silicon. Whether the $4 billion price is worth that bet is the question the company’s board is now weighing.
